HERITAGE: Many would argue the humble Cmoy is the headphone amp that started a headphone DIY craze. It was originally designed by Chu Moy and I’m sure tens of thousands have found their way into Altoid tins and other DIY enclosures. It’s similar to the original Grado RA1 headphone amp and is extremely simple. It’s an ideal first time DIY project and there are completed Cmoy amps from a variety of sources—although many deviate significantly from the original such as the Mini3 does.
BUY OR BUILD? I could make a Cmoy from scratch, or use someone else’s PC board, but decided to save time and test a ready-to-rock version complete with a Mini3-like aluminum enclosure. The Cmoy tested here was purchased with Buy It Now on eBay for $39 with free shipping direct from China. It’s a classic dual battery design without any gimmicks or supposed “enhancements”.
FATAL FLAWS & eBAY: Nearly every “no name” audio or DIY product I’ve encountered sold directly out of Asia on eBay has been seriously flawed in one or more ways. This Cmoy was yet another example. I wonder if these flawed products were originally intended to be sold through regular retail channels before someone found the Fatal Flaw. Sometimes the flaws are relatively obvious, as is the case with this Cmoy, and sometimes they only show up if you make the right measurements. But, out of probably 20+ products I’ve encountered, I’d say 90% had at least one significant flaw like this Cmoy does.
NO RETURNS: Would you return a $39 amp to China? I suspect the vendors know few will and eBay's used to “liquidate” flawed audio products--sort of like a factory outlet store selling “irregular” clothing. The flaws are not disclosed, the vendors frequently change their names to dodge poor feedback, and returns are expensive or not allowed. The seller for this amp has disappeared from eBay listings.
WHO NEEDS GAIN? Say you have an iPod. And, like most iPod’s, it can only manage about 0.5 volts of output. That works great with typical 16 ohm portable headphones but falls short for your 250 ohm full size Sennheiser cans which need more like 2 volts. So, in this case, you need about 4 times or 12+ dB of gain. This Cmoy has 0 dB of gain. The 0.5 volts your iPod can manage is all you’ll get into your Sennheisers. Net gain is zero and the Sennheiser’s don’t play loud enough.
IGNORING THE FLAW (for now): I was curious how this amp would measure ‘as-is’ partly to see if it was even worth spending more time on. The output performance of the op amp was my biggest concern and that’s relatively independent of the modest gain headphone amps need. So I started making measurements and was impressed enough I ran a full set of measurements.
THE AMP: The quality is decent enough and similar to the assembled version of the Mini3. It’s a nice black anodized extruded aluminum enclosure with thumbscrews for easy battery access, metal front and back panels, silkscreening, etc. There’s even a metal volume knob and the volume control feels good. It’s externally much like the AMB Mini3 except there’s no DC power jack for charging the batteries. Unlike the Mini3, this Cmoy has DC blocking capacitors on the inputs to protect your headphones if your source has a DC offset. This also allows the volume control to work better.
HISS & NOISE: Having no gain helps a lot with noise. Even with my SuperFi IEM’s I couldn’t hear any hiss.
SUBJECTIVE SOUND QUALITY: Driving the Cmoy from the headphone output of my Benchmark DAC1 using a variety of headphones it sounded very clean with no obvious flaws of any kind. It did not, of course, make the signal any louder but the Benchmark has plenty of output so that wasn’t a problem.
MEASUREMENT SUMMARY: The Cmoy surprised me. The 4556 op amp doesn’t get much love relative to other more expensive op amps. So I wasn’t expecting amazing numbers even at 150 ohms and I expected it to really stumble into 15 ohms. It’s worth noting the zero gain configuration helps out a bit—especially reducing noise. But even taking that into account, this Cmoy delivered promising performance. It also shows a conventional 2 channel design, even with an inexpensive op amp, can outperform a more elaborate 3 channel or virtual ground design:
| Measurement | 4556 Cmoy | AMB Mini3 | FiiO E5 |
| Frequency Response | +/- 0.0 dB Excellent | +/- 0.1 dB Excellent | +/- 0.1 dB Excellent |
| THD 1 Khz 150 Ohms | 0.001% Excellent (1) | 0.002% Excellent | 0.005% Excellent |
| THD 1 Khz 15 Ohms | 0.003% Excellent (1) | 0.017% Good | 0.012% Good |
| THD 20 hz 15 Ohms | 0.005% Excellent (1) | 0.01% Very Good | 0.6% Poor |
| THD 20 Khz 15 Ohms | 0.02% Excellent (1) | 0.45% Poor | 0.05% Excellent |
| IMD CCIF | 0.003% Excellent (1) | 0.043% Fair | Not Measured |
| IMD SMPTE | 0.003% Excellent (1) | 0.009% Very Good | 0.006% Excellent |
| Noise (ref 400 mV) | -96 dB Excellent (1) | -94 dB Excellent | -86 dB Fair |
| Max Output 15 Ohms | 67 mW Very Good | 104 mW Excellent | 108 mW Excellent |
| Max Output 150 Ohms | 180 mW Excellent | 38 mW Fair | 22 mW Fair |
| Output Impedance | 0.1 Ohms Excellent (1) | 0.9 Ohms Very Good | 0.7 Ohms Very Good |
| Crosstalk 15 Ohms | 68 dB Very Good | 40 dB Poor | 46 dB Fair |
| Channel Balance Error | 1.1 dB Fair | 1.14 dB Fair | Sample Problem |
- These measurements are unrealistic for gains greater than 1X
PLEASE DON’T SHOOT THE MESSENGER: Some might not believe a $39 pre-assembled Cmoy with a $0.50 op amp can measure this good. Before sending me hate mail or posting cranky comments please read my article about subjective vs objective audio. The schematic is posted below for the amp I tested. It should be easy for another engineer to verify these measurements, and indeed, nearly all of them are consistent with the JRC 4556 datasheet.
BOTTOM LINE: I was impressed the 4556 op amp can deliver this kind of performance into low impedance loads. I do, however, need to modify the Cmoy so it has some gain and re-check the appropriate measurements. Check back for those updates. But assuming most of the performance is similar it would be a high performance bargain for someone wanting a portable amp.
TECH SECTION:
CIRCUIT DESIGN: The schematic (from my simulation software) is shown below with the component values of the eBay Cmoy as I received it. The gain is (1 + 470/100000) or pretty much a factor of 1 aka 0 dB or unity. The use of a 47K potentiometer is also not ideal as it will have much higher Johnson Noise than a 10K pot would—but with no gain that doesn’t matter much. For comparison, classic Cmoy schematics can be found at Tangentsoft and Headwize.
GOLDEN EAR OP AMPS: The preferred op amp for Cmoy designs seems to be the TI/Burr-Brown OPA2134 or OPA134. Tangent has an op amp list that describes trade offs and even supposed sound quality. The 4556 doesn’t even make that lengthy list. Cmoy op amps seem to be sometimes chosen based on heavily biased sighted listening tests and an apparent misunderstanding of what specs matter for audio. The greatest single weakness when using an op amp as a general purpose headphone amp is current capability. The OPA2134 is only rated for half the current output of the 4556—35 mA vs 70 mA. In fact many (most?) op amps on the Tangent list are rated for less than 70 mA.
SLEW RATE MYTH: There’s a myth faster slew rate is highly desirable and some DIY sites and forum members throw around impressive slew rate numbers. But a headphone amp only needs about 1 V/uS of slew rate to handle any realistic signal it will ever see. So a 30+ V/uS op amp is no better than the lowly 4556 rated at 3 V/uS. Faster op amps may perform worse in other areas—especially power consumption which is important in a battery powered device.
SPEED OVERDOSE: It’s easy to understand how a “fast amp” seems desirable but think of it this way: If a single dose of aspirin gets rid of your headache using 100 times that amount is generally a bad idea as the side effects could be lethal. The same is true of ultra fast amps—they’re often less stable, more power hungry and noisier. Best-in-class designs are all about understanding the big picture, what matters most, and balancing the most important parameters and trade-offs. Even with a cost-no-object design, slew rates well beyond what’s needed are often undesirable. This is a long way of saying the 4556 is plenty fast in this application.
4556 VS OPA690 VIRTUAL GROUND: As mentioned above, the 4556 datasheet specifies 25 volts p-p into 150 ohms for 83 mA of peak current. That’s 166 mA total current from both channels. The OPA690 in the Mini3 is rated at 160 mA and shared between the channels. In other words, on paper, the two designs have similar peak current capability. In reality, as you’ll see below, the 4556 in this Cmoy far outperforms the Mini3 in nearly all measurements. This is largely because the Cmoy uses a conventional “real” ground vs the virtual ground (“third channel”) in the Mini3.
NO SERIES OUTPUT RESISTOR NEEDED: The 4556 is supposedly current limited. And, indeed, I tried to blow this one up but failed. It measures just as great after driving a short as before. The series short circuit protection resistor required by the AD8397, and placed inside the feedback loop in the Mini3, is a likely reason for some of its poor performance. The 4556 can be directly connected to the headphones as it is in this design.
CMOY PCB: In the photo you can see the 470 ohm and 100K 1% feedback resistors flanking the JRC 4556. The two 100K resistors above them in the photo provide the input bias for the op amp. The LED dropping resistor is in the top left. Blue boxed film capacitors are used for the input caps (bottom of photo) and decoupling (flanking the 4556). The volume pot is a 9mm enclosed type with a metal cast bearing. The 4556 is soldered in place making “opamp rolling” more difficult.
BATTERY LIFE: The idle quiescent current for the 4556 is rated at 8 mA and that’s exactly what I measured. The power LED, however, only runs from one battery and adds about 2 mA. From a 200 mAH battery this is 20+ hours of battery life at low listening levels and perhaps 10+ hours if you have power hungry headphones and like it loud. This is roughly triple the battery life of the Mini3 as there’s no virtual ground and the slower 4556 consumes less power. The LED is a bit of a design flaw as it may help one battery die before the other by adding 20% more idle drain to one battery—see Uneven Battery Life below.
UNEVEN BATTERY LIFE : Several people, such as the author of this Tangent article, claim dual battery designs may be a poor choice because if one battery runs dead before the other one. The theory is the amp will output DC and may damage your headphones. I discuss this in more detail in the Tech Section of the 3 Channel Virtual Ground article. In testing the theory, this Cmoy still has only 4 mV (negligible) DC offset with one 9 volt battery fully charged and the other breathing its last breath at 1.3 volts. As battery hits 1.2 volts, if it’s the negative battery, the amp just dies with no DC offset. If it’s the positive battery dying the amp will sometimes output serious DC—not a good thing. Long before it gets to 1.2 volts, however, it clips the audio signal badly and sounds terrible. So, in other words, anyone listening would have ample warning it was dying. Even single battery amps can output DC when the battery drops too low.
BATTERY MITIGATION: If I were using this amp on a daily basis, I’d disconnect the LED so the power draw was the same from both batteries, and use brand new alkaline batteries from the same package (or reasonably matched rechargeables). The batteries will more or less die at the same time and any DC problem at the very end when they’re down to the 1.3 volt threshold will quickly finish them off. Your headphones should be relatively safe—especially if you’re listening when it happens. I would , however, suggest not leaving the amp on when you’re not listening to it. But that’s true even of a single battery amp. The only 100% cure is a DC protection circuit.
DC OFFSET: I measured 4.5 mV in both channels which is sufficiently low. Contrary to myth, it doesn’t change with even large battery voltage mismatch (see above).
FREQUENCY RESPONSE: It doesn’t get much better than this. It’s dead flat from 10hz to 48 Khz. The slight 0.1 dB roll off you see at 5 hz are the 2.2 uF input caps. The channel balance is near perfect but with the volume control all the way up and no gain, that’s not surprising (although my FiiO E5 proves otherwise). This is driving 150 ohms to my standard reference output of 400 mV RMS:
POWER OUTPUT: Cmoy amps are intended for higher impedance headphones. This one does the job with a very healthy 180 mW into 150 Ohms. By comparison, the Mini3 managed 38 mW into the same load. Into a 15 ohm load it can manage 67 mW--a lot better than most portable players including any iPod or even the Cowon player I tested. Into 600 ohms you would get a still healthy 50 mW. To put this in perspective the popular Sennheiser HD600/650’s need about 2 volts RMS for any sane person and the 4556 with dual 9 volt batteries can deliver over 5 volts (nearly 15 volts peak-to-peak) to those headphones. The 15 ohm distortion is surprisingly low at about 0.003% at lower levels and it stays below 0.008% right up until it clips. The 150 ohm distortion is also seriously low in the 0.0008% range around 1 volt output:
POWER OUTPUT 33 OHMS: Because a Cmoy isn’t designed to drive 16 ohms, I also tested it at 33 ohms where it managed a very respectable 121 mW or roughly the same as the Mini3. And this is with both channels driven:
THD vs FREQUENCY: Here’s the THD+N at 400 mV output into both 150 ohms (yellow) and 15 ohms (blue) versus frequency. The drop above 10 Khz is due to the harmonics falling above the audio band. The performance is excellent into both loads and consistent with the 4556 datasheet:
THD SPECTRUM 150 OHMS: Not much to see here. Both the 2nd and 3rd harmonics are way down at –110 dB. My rule of thumb is anything below –70 dB for the 2nd harmonic and –80 dB for everything else is very likely inaudible. So this clears the bar by a mile:
THD SPECTRUM 15 OHMS: Making it work harder, here’s the result into 15 ohms. The relatively benign 2nd harmonic is still way down at –90 dB and the 3rd harmonic is at nearly 100 dB. This is way better than I was expecting into such a tough load:
RESIDUAL DISTORTION: Here’s the residual distortion waveform at 400 mV out into 15 ohms. It’s clean with no surprises. The 4556 is a very decent op amp even with a 15 ohm load (most op amps are only specified down to 600 ohm loads, the 4556 is specified down to 150 ohms but 15 ohms is another order of magnitude worse):
THD+N 20 HZ: Into 15 ohms at 20 hz the Cmoy does very well. The benign 2nd harmonic is just over –90 dB and everything else is below –100 dB:
THD+N 20 KHZ: This is also an excellent performance into 15 ohms (measurement bandwidth to 80 Khz):
CCIF IMD: Another great performance into 15 ohms. Everything in the audio band is down around –90 dB or better:
SMPTE IMD: Again, great performance. The sidebands around the 7 Khz signal are below –100 dB:
INTERNCHANNEL IMD: The channel shown is driving 0.9 volts at 1 Khz into 15 ohms. And the other channel is driving 0.9 volts at 300 hz into 15 ohms. I normally run this at 1 volt but that’s clipping for the Cmoy, so I had to lower it slightly to 0.9 volts. The result here is further proof how much better a “real ground” performs than a virtual ground. If you look at this test on the AMB Mini3 the result was far worse (see 2nd graph below). Here there’s mostly just the expected 300 hz crosstalk and the expected harmonics of the 1 Khz signal. There’s almost no actual IMD despite this being a Class A/B op amp and all the power supply ripple and “ground contamination” the 3 channel proponents claim are a problem:
MINI3 INTERCHANNEL IMD FOR REFERENCE: Compare the Mini3’s virtual ground design to the result above and note the much higher intermodulation products and dense “forest” of distortion created by the shared artificial ground:
VOLUME & GAIN: This Cmoy, oddly, has unity gain with the volume all the way up. In other words it has no voltgae gain. This is the design’s Fatal Flaw:
OUTPUT IMPEDANCE: With a 100K “load: the voltage was 401.4 mV (see above) and dropped to 398.5 mV with a 15 ohm load. This works out to an extremely low output impedance of 0.1 ohms (which is at least partly my cables, the jack, etc.). The reason it’s so low is there’s no series output resistor or capacitor between the op amp and the headphone jack and the op amp is operating at maximum feedback. This is as close to ideal as it usually gets.
CHANNEL SEPARATION (CROSSTALK): Again, this shows how much better a real ground performs compared to a virtual ground. This is excellent crosstalk performance and is in the range of just the output jack itself. The ground layout on the PCB helped here:
NOISE: The noise performance at half volume was excellent. But there’s a reason—no voltage gain. Most headphone amps have at least 8 - 12 dB of gain but this one has none. It shows the 4556 itself is very quiet:
CHANNEL BALANCE: The worst case channel balance was, as is often the case, at –45 dB--the lowest volume setting I measure at. It was only slightly over 1 dB which is very respectable and typical of the better analog pots (NuForce should take notes here!):
PHASE: Like the Mini3, the Cmoy had essentially perfect phase performance:
SQUARE WAVE RESPONSE: Driving the reactive load of real 16 ohm headphones (Sennheiser CX300’s) the Cmoy had only a tiny bit of overshoot and no significant ringing. The tiny peak could be removed by adding some compensation to the feedback loop. Overall this is excellent performance:
SLEW RATE: The large signal performance is shown below and the slew rate measured 3.8 V/uS which exceeds the worst-case 4556 spec of 3 V/us. The Cmoy only needs a 1.2 V/uS slew rate so this is plenty fast enough:
THE FINE PRINT: The measurements, unless otherwise noted, were made at max volume which is unity gain. The batteries were 8.4 volt Ni-Mh rechargeables and measured 9.0 volts while running the amplifier. All measurements were consistent with my more recent reviews and the methods outlined in my Testing Methods article.
TECH SECTION SUMMARY: As I said in the first part of the review, I was impressed how well the 4556 works as a headphone amp. Considering the measurements are basically of the 4556 itself, with little besides potentially the PCB layout and batteries to hinder its performance, most of this shouldn’t be a huge surprise. But I was surprised the power output matched the Mini3 into 33 ohms and the distortion was so low into 15 ohms. Whoever designed this Cmoy PCB did a good job with the layout, grounding, etc. The measurements clearly show the superiority of a real ground vs a virtual or “3 channel” ground. I need to correct the gain flaw and conduct some follow up measurements, but consider the 3 channel/virtual ground myth further busted. And don’t underestimate a lowly $0.50 op amp or Cmoy!
Although aren't most cmoys built with audiophile-approved OPA2132/4 op amps? I suspect they may not fare as well in some respects as the high-output-current 4556 in this application.
ReplyDeleteWhile I'm thankful for the time you invested in those very interesting measurements, I've to say I'm a bit turned off by the at times excessive rhetoric of the article.
ReplyDeleteThe NJM4556 is not considered "garbage" on diy audio forums. It's the opamp used in the well known Grado RA1 and many amps have been built by diyers with that opamp.
If the opa2134 is often chosen for cmoy, it's mostly because it's a jfet input opamp, making things easier on newbies. It's also widely available, while the NJM4556's supply is more erratic.
How about this one?
ReplyDeletehttp://cgi.ebay.co.uk/ws/eBayISAPI.dll?ViewItem&item=170647383588
Would I get similar performance as the one above?
Thanks
Chris
00940, thanks for your comments. Point taken on the rhetoric--I'm still getting beat up by the subjective "believers" and I'm letting it affect the tone of my writing. I'll do some more editing.
ReplyDeleteGood point about the Grado RA1 using the 4556. I revised the 4556 comments a bit. I wasn't trying to say everyone thinks the 4556 is garbage but I've seen several threads where someone swapped out a 4556 with something more exotic and claimed big improvements. Many also claim they're "too slow". And the 4556 doesn't even make the Tangent Op Amp List.
Thanks again for your comments.
Dark, the particular eBay Grado clone Cmoy you linked says it has a rail splitter which makes it a bit different. I'm not sure if they're using the rail splitter as a virtual ground, or just to try and prevent the "uneven battery discharge problem".
ReplyDeleteTo Anonymous asking about the OPA2134 versions, you are correct they can't deliver as much current. I talked about that in the first part of the Tech Section. I think the 4556 is a better op amp for headphone use.
I think you should try the resistor in the output (but within the feedback loop). You seem to have reservations about that, but this trick can actually improve things when you have reactive loads. And when the output impedance is very low, such as 15 Ohms, the OpAmp output has an easier job.
ReplyDeleteOtherwise: Good stuff, I like it!
The 4556 opamp has obviously been a standard headphone driver in CD players for 20+ years for good reason. It certainly isn't rail to rail or anything, but given some decently high supply voltages it's doing a pretty good job.
ReplyDeleteFunny to see the Grado RA1 brought up. It's a prime example of paying for a fancy housing with about the cheapest electronics that'll do the job.
Here's a few things that came to mind when reading the article:
1. "No gain". Make that "no voltage gain". A unity-gain amplifier (buffer, voltage follower) can have lots of current and thus power gain. A look at input and load impedance should make that clear. So much from the nitpicking department. ;)
This can be quite relevant in practice. Take a number of "Euro-capped" MP3 players which use approximately 47 ohms of output resistance to drop volume on low impedance in-ears, for example. Suddenly even a buffer has an effective voltage gain!
2. Your 9V rechargeables are likely to have lower output impedance than conventional alkalines, let alone cheapo "heavy duty" batteries. Thus low impedance driving with non-rechargeable types is expected to be worse.
These 9V blocks use AAAA cells (no typo there), and from what I remember you're lucky if you get 200 mA out of them. (1/10 the capacity and output current compared to an AA cell, seems quite reasonable.) Hence why you couldn't get the OP to burn up.
3. The interchannel IMD and gain plots have odd vertical axis scaling.
4. In terms of gain, I would aim for about 2 or 3 (6 to 10 dB). A gain of about 3 is a typical value for a Cmoy, and usually adequate for medium to low impedance loads. It's a nice touch that the designer bothered to include a gain-setting resistor. Maybe they merely got the value wrong by a factor of 1000 (could have been during assembly, too)...
5. Expect noise and distortion to vary with volume pot position (worst at about -6 dB). Something worth playing around with if you feel so inclined.
6. No output series resistor, hmm. On the one hand, this is good news for problematic in-ears, however it potentially makes the amp susceptible to capacitive loading. How many nF does it take without breaking into oscillation? It is not unusual to find 3m headphone cables with 1..2 nF.
7. Using a 20 kHz lowpass when taking THD vs. f may be common practice, but in my opinion it's utter nonsense. It does allow you to see which component is dominant, which can be handy sometimes. However, as you point out quite correctly, this dropoff isn't real at all - associated IMD (another side of the same nonlinearity) doesn't care one bit.
When you specify an amplifier with THD from 20 Hz to 20 kHz, using a 20k lowpass essentially amounts to cheating.
8. I would certainly like to see a more advanced harmonic distortion metric than THD put to practical use (say, for THD vs. f). In spite of the first empirical weighting methods dating to the 1930s, that would be pretty bleeding edge. It'll never be possible to do it exactly since transducer gain always figures in (you'd have to go with a worst-case estimate for a general purpose amp), but maybe there was some reasonably simple method that jumped at you when going through these AES papers? Anything is better than nothing, really.
Thanks Stephan and also the previous anon comment. Good stuff!
ReplyDelete1 - I agree about voltage gain and I mention in the first part of the review the Cmoy has, in effect, potentially useful current gain.
2 - The battery impedance issue does come into play although it's mitigated at typical output voltages by both the power supply caps and the high PSRR. For a design that's voltage swing limited, like the Mini3, it's a big factor in maximum power into low impedance loads.
I agree battery impedance may a factor in how long the 4556 can survive a short. The JRC data sheet implies the 4556 has current limiting, and my tests confirm it seems to, but it may still eventually die from excessive dissipation. In a headphone amp the need is primarily to survive brief shorts.
3 - Thanks for catching the weird scaling. The data is 100% valid but the dScope software will do the math if you don't let it choose how many divisions are used.
4 - I think 3X - 4X gain is the most useful. If you subscribe to the "HD600's need 2 V RMS" theory that means you need 4X gain with most portable devices. The Mini3 is 5X by default which is a bit high--especially given its limited output swing of < 3 V RMS. That also is hurting its distortion performance--especially at high frequencies.
5 - I typically check various parameters at different volume settings. One nice thing about the dScope is you can see most any parameter in real time--THD, crosstalk, noise, channel balance error, etc. So it's easy to just turn the knob and watch the value. In my experience, especially with the Mini3, its poor crosstalk, and high 15 ohm distortion, swamped the results so the volume control had no effect. The Johnson Noise, however, can be significant in some designs. I'll try to include more comments on that in the future. It's also a trade off between measuring "best case" performance and "real world" performance that's more typical of how it's typically used.
6 - I need to further check stability without a series resistor but I did run it into real headphones with a 10 Khz square wave and the result is shown--it was better than the Mini3 with no ringing and only a hint of overshoot. There are other ways to solve that problem besides a series resistor.
7 - That's why I run my 20 Khz THD measurement out to 96 Khz. But to be consistent with Stereophile and other "commercial" reviews, I usually limit sweeps to the default 22 Khz. I also think the CCIF IMD test is generally more revealing of audible HF problems.
8 - I've been working on a blog article that tries to bring together some of the current body of knowledge regarding various distortion and non-linearity measurements and sound quality. It's a difficult topic, and to be credible, requires lots of references. There has been some relatively recent articles published that shed some new light on a few areas.
Re: 8, "It's a difficult topic":
ReplyDeleteYou bet it is! It's like a big puzzle waiting to be put together.
Essentially the goal is determining all the nonlinearities our hearing exhibits (magnitudes included), then you'll know what they mask. Given the effectiveness of psymodels in lossy coding, we can't be that far off from knowing all the contributions, but like I said things still want to be put together and connected to elements of the auditory system.
What fun. :)
This blog is amazing and the only I'd read, and have been doing so since your first entries.
ReplyDeleteWhat devices do you see yourself testing in the future? One of particular interest to me would be the Cypher Labs AlgoRhythm Solo, as it's often referred to as a game-changer.
Keep up the good work!
Thanks for the kind words. I'm not sure if you want the AlgoRhythm Solo tested because you think it's great or a waste of money? I suppose it depends on which iPod, but the ones I've tested already have very capable DACs in them and measure very well (see my Clip+ review). And Apple is supposedly switching to a 24 bit Wolfson DAC in future models. You could buy a whole new 24 bit iPod for less than half the cost of the Solo.
ReplyDeletePersonally, I think a portable battery powered DAC with no headphone output is not a very useful product. There's no technical reason an amp cannot be built into the device (much like the FiiO E7). But, instead, you're expected to carry around 3 portable devices to listen to your music?
And the cost of even the least expensive AlgoRhythm Solo/Amp pair is several times what an iPod costs. It would be fun to do a blind test with some very revealing IEM's of a "raw" Touch 3G and running through the Cypher Labs pair.
I flinch everytime someone mentions something making a "night and day" difference, and the AlgoRhythm Solo is a device that's been said to do just that by a great deal of people. Jude even covered it himself in his video blog.
ReplyDeleteA blind test versus any 3rd or 4th generation Apple device is likely to show that differences, if any, are subtle at best. Tests could potentially identify it as a fraud, or atleast put the high asking price into perspective.
b0ck3n Thanks for the comments. Especially: "Jude even covered it himself in his video blog"... ALO is a prime Head-Fi sponsor selling the AlgoRythm. I'm sure his review is entirely unbiased and he had no other motives for the video review being front and center on the Head-Fi home page... ;)
ReplyDeleteGood job, NwAvGuy, I really appreciate your work.
ReplyDeleteYou don't have an alternative name, do you? Something like John or Peter, or Stewart? NwAvGuy seems so impersonal, and hard to pronounce!
You want AlgoRhythm measurements? Look no further than here. I know, RMAA and absolute levels, but for an idea of performance it should be OK. Unsurprisingly, I see no indications of any "night and day" type differences, or even better DAC performance compared to the 4G Touch.
ReplyDeleteRegarding handles, I'd have to agree that "NwAvGuy" is a veritable tongue twister indeed. On the other hand, it's not a bad idea to stay as anonymous as possible when you've potentially got an angry mob of subjectivists following you.
Well, maybe Sinéad O'Connor has the right kind of advice. :p
Thanks for sharing that treasure, Stephan. That the Solo performs notably worse than the 4G Touch is astonishing. It's just a shame that the iPods can't seem to shake that high outputZ.
ReplyDeleteI haven't personally measured an iPod Touch 4G or an iPhone 4 but based on the measurements of others, it appears Apple lowered the output impedance down to 2 ohms or less. That's arguably low enough. I'm waiting for the new 24 bit Touch 5G and I'll be testing one of those to see if the 24 bit hardware makes any meaningful difference. I'll also report the output impedance.
ReplyDeleteEven so, in the measurements linked to by Stephan the same DAC (Touch 4G) messes the TF10s FR up pretty badly, but coupled with a <1 Ohm output impedance amp it's pretty much dead flat. Also, by ear, I'd have to say my iPhone 4 puts some bumps and troughs into the FR of my ES5, that the Clip+ doesn't (I won't exclude the possibility of expectation bias coming into play).
ReplyDeleteVery much looking forward to any and all reviews down the line!
b0ck3n yeah I just looked it's about 3.2 dB of variation with the TripleFi on the Touch 4G. I don't know how wild the TF-10 impedance is. Looking at the NuForce uDAC-2's 4.4 dB of variation, which I know is around 6 ohms, the Touch 4G does better implying the 4G's impedance is a bit below the 6 ohms of the uDAC-2. I'd say around 4 - 5 ohms without doing the math.
ReplyDeleteI also have to point out those are all RMAA measurements and RMAA is prone to all sorts of problems. See my RMAA Article for more. In this case, however, at least the frequency response should be reasonably accurate.
The other interesting detail is the Leckerton UHA-4, which Leckerton told me had a 10 ohm output impedance, measures nearly flat in those tests. Leckerton did supposedly lower the output impedance after I published my NuForce review. I don't know when the Markus Kraus testing was done but, if its correct, it does confirm Leckerton fixed the UHA-4 which makes it an attractive reasonably priced option to the E7.
A simple amp like this plus a DAC that can output ~2 Vrms .. would be a wonderful DIY project.
ReplyDeleteHave you tried adding gain? My guess is that the DC offset would increase quite a bit. If that's the case what do you think about moving the gain into an additional opamp that sits right before the volume control. Is something like that feasible/does it make sense?
I have added about 12 dB of gain and re-checked several measurements. I just need to write up the article. As expected, the distortion, noise and offset all increased but the overall performance is still respectable. It's also still very stable.
ReplyDeleteAdding a stage would help in several ways. I talk about some of that in the reviews (my front wheel drive car analogy). Asking one stage to do everything is a compromise but it helps battery life and keeps the size and price down.
A combined DAC/headphone amp DIY project is another can of worms. I keep hoping for a better DIY-friendly USB option besides the ubiquitous (and relatively ancient) TI PCM2xxx ICs. But, that aside, if you want portable battery power with USB support, it's hard to beat something like the FiiO E7 or Leckerton UHA-4 on size or price. If you mean a desktop USB Headphone DAC that's a bit more practical--something like a NuForce uDAC-2 without the serious flaws.
Figured I'd post some iPod output impedance numbers for future reference.
ReplyDeleteTouch 4G - 7 ohms
Classic 160 - 5 ohms
Nano 6G - 2 ohms
iPhone 4 - 0.9 ohms
Thanks b0ck3n. I'm curious where the Touch 4G number is from? A fellow in Germany who seems to know what he's doing insists it's less than 3 ohms. I haven't tested one myself so I don't really know.
ReplyDeleteThe review here specs it at 7.2 - http://www.kenrockwell.com/apple/ipod-touch-4g.htm
ReplyDeleteLooking at Markus Kraus' measurements, 3 ohms seems accurate as the Touch curve swings less wildly with the TR10s than the Classic at 5 ohms. I suppose he came up with the figure - does he still operate a website?
I'll add iPad at 2 ohms and iPad 2 at 0.9 ohms. Let's hope Apple's next line of iPods keeps the positive trend going. You'll be the first to let us know, right? ;)
Hey NwAvGuy,
ReplyDeleteI was wondering if you have heard of / have any oppinions on "The Wire" headphone amp over at the DIYAudio forums. Some really impressive measurements have been shown and it seems like a simple no BS design.
Here is a link:
http://www.diyaudio.com/forums/headphones/179298-wire-ultra-high-performance-headphone-amplifier-pcbs.html
Anon, I think The Wire is a solid design and I like the LME49600 a lot. I also like the fact he ran Audio Precision measurements to prove the performance--that's very rare in DIY. On the downside I think The Wire is more expensive than it needs to be. It also uses surface mount parts which are a challenge for some people.
ReplyDeleteI started a DIY audio thread regarding a more simple design ( http://www.diyaudio.com/forums/headphones/186188-national-lme49600-reference-design-project.html ). We got a bit off track in the thread, but I'm still researching similar designs. My goal was a reasonably priced, no BS design, that's easy to build and performs well. One possibility is the Sojstrom QRV09 and I'll be publishing results for it soon. But it also has the downside of being mostly surface mount and has one very difficult component to properly solder (the TI TPA6120). It also suffers from the TPA6120's 10 ohm output impedance requirement but I'm researching ways around that.
If you continue on with that LME49600 design I think it would be great if you used a couple buffers per channel for maximum current output for the planar magnetic designs that are becoming more popular these days, since they're relatively low impedance and low efficiency.
ReplyDeleteMaverickronin, I am still researching options. My testing of the FiiO E9 and Sjostrom QRV09 have pretty much ruled out using the TPA6120. Do you have some current examples of popular planar magnetic headphones that are difficult to drive?
ReplyDeleteHiFiMan (their headphones are a lot better than their players)
ReplyDeleteHE-4 38 ohm, 86db/mw
HE-5 25 ohm, 87db/mw
HE-5LE 38 ohm, 87.5db/mw
HE-500 38 ohm, 89db/mw
HE-6 50 ohm, 83.5db/mw
Audeze LCD-2 50 ohm, 91db/mw
Fostex T50RP 50 ohm 98db/mw
(subjectively the manufacturer's rating seems optimistic to me compared to other 'phones I've owned but I've only got a cheapo multi-meter and spl meter from radio shack so I can't exactly verify this myself)
There are also lots of older ones that have similarly low sensitivity but impedance of more than 100 ohms or so which makes it more a question of gain and the limits of the power supply rails rather than raw current output.
I think in the spirit of making it a "one size fits all" amp it would be nice if it could drive any of those cleanly to ear splitting levels after factoring in some headroom for EQ, but I don't know how much complexity of cost it would add.
They're also rated to play very loudly and handle a lot of power. HiFiMan does't give max power ratings for their models, but the LCD-2s are rated for 15 full watts and my T50RPs are rated for 3 watts.
Thanks Maverickronin. I know the LCD-2's have a reputation for being challenging to drive and the HiFiMan planars look even worse on paper. But, that said, things are not quite as bad as some might think. While the Mini3 would fall on its face into any of the headphones above, you don't have to spend a lot to properly drive them.
ReplyDeleteThe Cmoy reviewed above can manage about 100 mW into the 38 ohms HiFiMan cans or about 107 dB for an 87 dB/mW sensitivity. That's a bit shy of the 110 dB many consider necessary to reproduce peaks at "live" levels of highly dynamic music but still fairly loud.
The FiiO E9, however, can manage about 800 mW into 38 ohms which, if the HE-5LE can handle it without compressing, would be good for an extremely loud 116 dB SPL.
The Cmoy would manage about 180 mW into the 50 ohm LCD-2 that's 114 dB which should be plenty loud for most. The FiiO E9 would deliver 600+ mW into the LCD-2 which is about 119 dB SPL.
Of course, all of the above assumes the manufacture's sensitivity and impedance ratings are accurate. It would be interesting to do a peak level test using a scope and real music with say the HE-5LE cans running off the Benchmark. But that gets into my subjective idea of "loud".
I know they aren't *quite* as demanding as some people seem to think, but it always seems disconcerting when I have to crank the volume that high.
ReplyDeleteI'd probably make some stupid error in the math, but you can see the real impedance as well as the voltage and power needed to reach 90dB SPL in the data sheets here.
http://www.innerfidelity.com/headphone-data-sheet-downloads
Thanks. Per Tyll's measurements the HE-5LE is not as sensitive as claimed at 41 ohms and 4.7 mW for 90 dB SPL.
ReplyDeleteThis Cmoy can manage about 180 mW into 41 ohms. So 10*LOG(180/4.7) = 16 dB. 16 + 90 = 106 dB SPL from the Cmoy. That's significantly under the 110 dB target.
The FiiO E9 can manage about 5.7 volts into 41 ohms or 800 mW. That gives 112 dB which should be loud enough for most tastes.
-12dB or so for my movie EQ though. I don't know how commonly other people do that sort of thing, so you may not want to design for it.
ReplyDeleteI'm not sure how relevant it is to mine, since the damping schemes are different, but one of the modded pairs of T50RPs on Innerfidelity shows 1.23mW for 90dB SPL when the manufacturers specs are 98dB/mw.
Maverickronin, there's a difference between maximum output (clipping) and gain. In the SPL numbers above I'm talking about what the amp is capable of with sufficient input signal.
ReplyDeleteThe "movie EQ" issue is a gain issue and the gain can, as discussed elsewhere, usually be adjusted prior to the headphone amp. The only time it would be a problem is if the driving source clips (or otherwise is limited) before the headphone amp reaches maximum output. For example, if your PC's line out can only manage 1 volt RMS, and the Cmoy has a gain of 4X, you're only going to get a max of 4 volts RMS out. Into the 41 ohm HE-5LE's that's plenty as it will clip around 3 volts or less. But if your PC can only manage 0.5 volts RMS out, you're only going to get 2 volts into the headphones no matter what you do.
So as long as the source can drive the headphone amp into clipping, you can likely add gain in the digital domain as needed to overcome EQ or "quiet" movie soundtracks.
The problem is that adding large amount of EQ has the effect of increasing the dynamic range of the sound, and movie soundtracks already have lots of dynamic range to begin with. If I have a 12dB "boost" centered on 20hz it actually makes everything else quieter because its all done in the digital domain and it can't actually go over 0dBFS, right?
ReplyDeleteIf I boost the master volume digitally it will take those 20hz peaks over 0dBFS and clip digitally, right? That's what it sounds like when I try doing that. At least I can tell that something is clipping somewhere.
That's the gain related part, but it also relates to maximum power too. (I think)
Say I'm listening to some EDM or trance (I don't use that much EQ with most other music) and I'm using the same +12dB at 20hz and say my "crank it for *one* song" volume is 100dB peak in the mid range and lower treble where human ears are most sensitive and you will actually lose your hearing if you listen too loud. That means the bass hits which won't damage your hearing (and thus can be much louder) will peak at at least (depending on how compressed the dynamic range was in the original recording) 112dB which may be well past clipping with some of the less efficient 'phones.
Right now my current amp provides just enough power for me to use those aggressive EQ curves w/o audible distortion and still get the volume I want, but the T50RPs are some of the more efficient planars currently on the market and I might end up getting something even less efficient later so I'm hoping to "future proof" my next amp purchase or build.
Does any of that make sense or am I misunderstanding things?
If you're increasing the dynamic range, then you're correct you might want more maximum power available to handle higher peaks relative to the average volume. But there's a prudent limit.
ReplyDeleteEQ by itself doesn't really change the overall dynamic range although, as you point out, if the bass is the loudest thing on a recording, and you boost the bass with EQ, then in effect you've altered the peak to average volume ratio possibly increasing the perceived dynamic range.
But the peak SPL calculations remain the same. If you dial in lots of bass boost, then the bass will clip first (i.e. at 112 dB SPL or whatever). So your ears (assuming flat response in the headphones) are still getting the same SPL peaks regardless of the EQ. But if it's confined to the deep bass, perceptually, it won't seem as loud.
The reason I say "prudent" is long term exposure to even 85 dB, and relatively brief exposure to 120 dB, causes hearing damage. So I'm not sure it's wise to have an amp and headphone combo that can deliver much more than about 115 dB on peaks. The 110 dB rule of thumb I mentioned earlier is typically sufficient.
Human hearing is much less sensitive to distortion, including clipping, at very low frequencies. Subwoofers, for example, often have 10% or higher THD when driven hard but we still perceive the bass as relatively high quality. So clipping the loudest deep bass effects on a movie soundtrack with your headphone amp isn't usually a big problem.
But if it's the pounding bass line in trance music you're listening to, then I'd seriously worry about hearing damage at the levels we're talking about (110+ dB). So while I understand the desire to have enough "headroom" for lots of bass EQ, I'm not sure it's healthy over the long run.
As an aside, alcohol and high SPL audio gear is often a bad combination. Human judgment as to what's "loud enough" seems to go out the window if you've had too much to drink. I can't count how many times I've seen damaged speakers where the owner admitted "yeah I was kinda drunk and must have gotten carried away". Headphones might not be damaged, but when you go ballistic with the volume control and have gear capable of 110+ dB SPL, your hearing will almost certainly suffer permanently.
A few points:
ReplyDeleteThey aren't supposed to be flat after all that EQ, and that's not what I'm going for with movies or trance/EDM. With headphones there's a trade off between what *sounds* accurate and what *feels* accurate and for some things I think the feel is more important.
I don't usually listen that loud. Normally I listen fairly softly but occasionally I'll crank it fairly loud for just a song or two. I'll have to measure it next time. I was just throwing out numbers to illustrate my point. I just never had a problem driving my amp into clipping with any other headphones I've owned before so now I'm a bit paranoid. Probably too paranoid, as you can tell.
Most of the music I listen to doesn't have that much dynamic range but some movie soundtracks seem like they do. I wouldn't be surprised by an explosion in an action movie peaking 30-40dB above the dialogue after I've added +12dB at 20hz or something. I've never actually tested this though.
The hairs in your inner ear that pick up midrange and treble are much more sensitive and prone to burning out, causing hearing loss, than the ones that react to bass frequencies so FWIR pretty much any low bass that isn't enough to rupture you eardrums won't cause hearing loss. Correct me If I'm wrong.
I know that we're less sensitive to distortion in low frequencies, but I'm going by what I hear. It could be that when the bass clips my amp spews out distortion at higher frequencies as well, but I don't know and I lack the gear to find out.
What made you think I was posting drunk ;-)
Also, most 2 and 3 way BA IEMs are already sensitive enough to damage your hearing from many DAPs and just about any stand alone head amp if you're not careful or responsible so what's the big deal about adding even more power?
I wasn't accusing you of listening drunk, just making a more general point. And you're correct many BA IEM's can already be lethal to hearing. I was just trying to define some basis for what's "loud enough" in terms of peak SPL levels. I can understand those wanting even more but, personally, I'm not sure it's a great idea.
ReplyDeleteI was just kidding. That's what the smiley was for.
ReplyDeleteGood info on the 4556...is this the one that you'll be using in the O2? Thanks for taking the time to write all of this up...it's greatly appreciated.
ReplyDeleteIt is nice to see someone targeting audiophiles with sensible devices that don't cost a kings ransom to buy. The typical high end hi-fi equipment have prices so ludicrously high, its hard to take it seriously. High end headphone amps that would be overpriced even if they were made of solid gold. You could fill a bath with electronic components for the prices these companies charge. Your amp looks like a practical and sensible kind of device. All you have to do now is put it in a case made of American Swamp Ash and laminated with crocodile skins, and run some internal connections with cable an inch thick and you will have the high end market sown up!
ReplyDeleteThis is CMOY with OPA2604 x 2:
ReplyDeletehttp://www.ebay.com/itm/OPA2604-x-2-Headphone-Amp-kits-CMOY-/250842462148?pt=LH_DefaultDomain_0&hash=item3a676037c4
Do you think its sound not worse than 4556 Cmoy?
OPA2604 is a good amp for this schematic?
I would not design a Cmoy with any other op amp besides the NJM4556. Nothing else even comes close except perhaps the OPA551. The OPA2604 is only rated for 600 ohm loads or higher. The THD is only rated into 1000 ohms. That rules out nearly all headphones. And it has half the current capability of the 4556. It's a very poor choice to drive headphones directly. See my O2 Design Process article for more.
ReplyDeleteThe vendors on eBay trying to use "designer" op amps in headphone amps are just trying to be "fashionable" or "trendy" but they're actually making their amps perform much worse. They never publish meaningful test data and I doubt many ever even properly test their designs. If they did, they would know the 4556 is a much better op amp in that application. Actually if they even bothered to read the datasheets they would know that.
For more on op amp myths see:
Op Amp Myths & Facts
For some proof, and comments on the OPA604 (the single version of the OPA2504), see:
Op Amp Measurements
Thank you for help.
ReplyDeleteI know that Behringer's Micromon MA400 have NJM4556 for headphone output. Sorry for maybe silly question, but... did this guarantee low output impedance and good perfomance of this headphone amp?
Behringer's tech support said me:
"The MA400 uses one stereo NJM4556 op amp to drive the output and the manufacturer does not specify its own internal output impedance".
Most 4556 amps I've seen (which have been mostly minimalist Cmoys) don't add any output resistance to the 4556. The 4556 by itself is well under my 2 ohm rule of thumb. Only if the designer decides to add resistance would it be higher. Behringer, for example, adds 47 ohms to the output of the headphone op amp in the UCA202. But it uses a much less capable op amp than a 4556.
ReplyDeleteSo the real RA1 (and its true clones) uses a rail splitter? I assume it has no added output resistance? The layout on ampedup77's and fred_fred_2004 is obviously indicating they are this same RA1 clone that is distinctly different than the cmoy nwavguy found on ebay. I'm wondering how its performance compares. If anyone spots a DIY vendor on ebay that's selling the ones with the layout nwavguy found, post it here. I'd really like to get one and have them make it zero gain and remove the LED. My Emu 0204 already has a 2.07V output, so I don't need any extra gain. Nwavguy, what is the O2's performance with zero gain?
ReplyDelete@Reticuli, the O2 performs even better at 1X gain than the measurements I've shown at higher gains. It's main advantages over the dual battery Cmoy are:
ReplyDelete- Double the output current to better drive lower impedances.
- Double the output power dissipation (half the heat per op amp chip) with one IC per channel.
- Rechargeable batteries and AC power.
- Selectable gain. You could have 1X and another gain setting of your choice for other sources or headphones.
- Battery management to prevent headphone damage if one battery dies before the other or one becomes disconnected. The same circuit also protects the batteries against over-discharge and reduces turn on transients.
- The downside is the O2 is larger and the complete amp is about twice the price (in DIY form). But that includes rechargeable batteries and the AC adapter. Someone could build a mint tin battery-only O2 for less than the cost of the Cmoy.
AFAIK, the real original Grado RA1 used dual 9 volt batteries as well and not a rail splitter. I think there have been at least two versions, however. But I'm not aware of any using a rail splitter for a virtual ground.
O.k. So that means this RA1 clone on ebay isn't quite one. Thanks. Still waiting for my O2 to get built. The E7's good, but I'd like to input the Emu 0204's 2V signal into it, hence my interest in those ebay cmoys and the E6. The latter would probably be more useful in portable use once the O2 is done, I guess.
ReplyDeleteThe Cmoy I tested is fairly true to the RA1. There are other eBay "RA1" amps, however, that use ground splitters. The O2 is a significant step up from any of them including the actual Grado.
ReplyDeleteAnd the JDS cmoy?
ReplyDeletehttp://www.jdslabs.com/products.php?page=cmoy
The JDS Cmoy uses a 20 mA TI rail splitter which isn't up to the challenge with many headphones.
ReplyDeleteSo JDS cmoy and this are virtually the same?? Or there is a difference?
ReplyDeleteDoes JDS offer a dual battery Cmoy? The ones I'm aware of use a single battery and rail splitter which is very different. See: Virtual Ground Amplifiers.
ReplyDeleteJDS does offer a dual battery option now. It's their 2.03 version.
ReplyDeleteThis is a ridiculous article. All these supposed CMOY amps using an opamp are terrible driving a dynamic load. You can't just place a dummy resistive load on the output then hook a scope up, you have to use mics to get a signal from the cans themselves.
ReplyDeleteWithin it's current capability, the two opamps sound quite different directly driving a load, but if at all interested in what is suitable you would buffer either of them, at which point most of the sonic differences disappear and they both start sounding like the same superior solution.
The author makes sweeping claims including one of objectivity yet when it comes to his challenge about how opamps sound the challenge has no objectivity whatsoever, only an ASSUMPTION that others cannot hear a difference between opamps. In a blind test I and many others can definitely tell a difference, but perhaps our hearing is intact while the author's is not and this accounts of a lot of the denial and an attempt to only measure things in a way that supports a preconceived subjective assumption.
However it is not "sounds better", it is which sonic signature a listener prefers to synergize with their cans. Some like OPA2134 sound dark and and veiled, NE5532 a bit more raw and lively, AD8610 being more neutral.
If you can't hear that then subjectively you have no reason to spend more than necessary on an opamp UNLESS you also insist that it's also true that buffering doesn't sound different and ONLY THEN after two subjective perceptions can you attempt to fein objectivity that the only thing that matters is current rating and use skewd tests which emphasize current rating.
Guess what? Music gets better still with a combination of a fast opamp AND higher current rating using a buffer. Realign testing to mirror real world results and you will have your numbers to support what you subjectively suggest is someone else's subjective mistake. Instead all we have are "what if's", worth case scenarios of what might go wrong instead of accepting that yes, higher performance opamps perform better and need more care in design to do so.
The author is totally lost. Over and over the mistake is made that an opamp is transparent based on anything other than the obvious truth: They are not all equal driving a load and that is not just a matter of peak current rating, even an untrained ear can hear the difference in blind tests within half of the rated output power of the lesser current rated opamp.
To suggest the majority switch opamps around and can't come to terms if they sounded the same is madness. In audio forums I mention several times a day that changes people propose are not likely to sound any different, but in the case of opamps directly driving cans they most certainly do.
@Anon, comments like yours help promote myths that have no basis in objective fact. Perhaps you should read:
ReplyDeleteOp Amp Myths & Facts
And then:
Op Amp Measurements
And then relative to test vs real loads, I've discussed that topic as well. Techniques like audio differencing allow comparing the input and output of an amplifier while driving a real load using real music. Guess what, op amps are still transparent even under such conditions when they're operated within their design criteria. See:
Music vs Sine Waves
Indeed, I discuss when an op amp is operated outside its design criteria, as most directly driving headphones are, they can and do often sound different. So we agree on that point at least.
But the reason they sound different has little to do with their slew rate or most of their other specs. It also has little to do with their cost or how exotic they are. It has everything to do with how gracefully they deal with being severely overloaded--something you can't find in the datasheets.
You call me "lost" yet you fail to provide a single measurement or objective reference to back up your mythical claims. I've written two articles just on the topic of op amps and have referenced several other articles such as massive work of Samuel Groenig. I have measured a dozen or so op amps in an audio application and published the results.
Perhaps best of all, I've challenged anyone to a proper blind test to hear the difference between op amps in a gain stage. Not one person has come forward. Would you like to be the first? The challenge is in the "myths" article linked above.
Having been an OPA2134 fan (I changed all the op amps in my CD player to this chip) I was quite skeptical that the 4556 would sound as good as the OPA2134s. I bought some 4556s in horrible smd packages that would not even fit properly on my carriers and replaced the headphone drivers. The phones are 32R Sonys. The sound is definitely different - there is a lot more detail in the high frequencies and the sound seems better balanced. I found myself listening a lot longer than I had intended as the music was more involving and more relaxing.
ReplyDeleteI then thought that the 4556 might drive the line from the preamp to the power amps better too (5m of 90pF/m microphone coax through 100R) so I swapped out the OPA2134s there too. Sure enough the high frequency detail/apparent volume was there too - CDs sounded more like my old vinyl. Curiously my 10 year old did not tell me to turn the music down like he normally does.
So I am sorry there are no measurements but all I can say is that the 4556 sounds better than the OPA2134 when driving 32R headphones and moderate lengths of cable.
What really bugs me is that the frequency response of the 2 opamps in the low gain circuits should be ruler flat to well above what I can hear, yet I can definitely detect a difference in tonal balance between them.
I suppose 1 explanation for the tonal change could be that I am playing the music louder because the 4556 is not trying so hard and so I hearing with a more curved equal loudness contour...
@Steve, without doing a level matched blind test (something that requires two identical pieces of gear for comparing op amps--i.e. two identical headphone amps with a different op amp in each) you really can't draw any conclusions. It's been proven over and over we humans can't rely on our ears when we know what we're listening to. It's called sighted listening bias and it's very well documented. In one study people listening to the exact same track played through the same hardware overwhelming thought they preferred one over the other when they really sounded the same. Our brains invent differences where none exist. See: Subjective vs Objective
ReplyDeleteI don't really think it is as simple as level difference.
ReplyDeleteIs it possible that the frequency response of the amplifier driving a reactive load actually peaks then falls? I have some CDs that have an annoying sibilance on vocals and perceived lack of higher frequencies (muted cymbals) with the OPA2134, and with the NMJ4556 there is little sibilance but a lot more of the higher frequency stuff.
Is there some way I could measure the frequency response of the system using a UCA222 for example? It strikes me that the effect is so gross that even a not very sensitive audio analyser should be able to detect it.
@Steve, believe what you want, but until you do a blind test you really don't know what you're hearing. Look up the BBC McGurk Effect video on YouTube. Our ears can blatantly deceive us. Sibilance is more subtle than hearing "Fa" when the someone is really saying "Ba". Check it out.
ReplyDeleteTo answer your question, the dScope analyzer is capable of detecting even tiny differences and errors well below the threshold of human hearing. Generally even an overloaded op amp will not display frequency response problems in the audible range of frequencies. Your theory has no basis in fact.
When a friend walks in, listens, looks puzzled, then says "What have you done to the bass? The boom has gone. The sound seems better balanced.", I take it to mean that the change is audible. I was careful not to give any clue as to what was changed, yet he described pretty much what I was hearing.
ReplyDeleteAnd why not? The op amps are driving abnormal loads; 5m cable runs and feedback loops with 3k resistance, so we should expect some kind of distortion. And that distortion should be measurable. I would expect less distortion from NMJ4556 as it has better current drive. I would expect less distortion from the OPA2134 because it is 5x more expensive and has higher slew rate(?).
My ears say current drive wins over cost. Is that not what you would expect?
The fact that I don't know how that distortion manifests itself (does all distortion sound tinny or maybe something more subtle like some kind of masking?) does not mean WE cannot hear some effect.
Are you saying that the O2 measures better than the CMOY, but don't bother building it because you cannot hear the difference unless you do a blind test?
You can't have it both ways you know!
Steve
@Steve, the friend/wife/girlfriend commenting something different has also been thoroughly debunked as invalid. Our senses are just not that reliable however much we want to believe they are. It would be like walking into a restaurant and saying the chef has switched brands of olive oil because of the smell in the air. And you can't walk into a room and "hear" a headphone amp playing.
ReplyDeleteI'm not trying to have it both ways. The 4556 is designed to drive headphones and the 2134 is not. The 4556 can drive a much wider variety of headphones than the 2134. So it has some very real advantages--especially when paralleled and operated as buffer as it is in the O2 Amplifier.
I have said all along it's not valid to compare op amps that are operated well outside of their design criteria. See: Op Amp Myths & Facts. If you compare the 4556 and 2134 IN A PROPER APPLICATION (i.e. in a preamp, output buffer of a CD player, etc, not driving headphones) you're very unlikely to hear any difference in a proper blind test. Both will be transparent.
Ok this is my last comment on this. I compared 2 Denon DCD825s last night. Mine has the 4556 with locally regulated power supply and direct coupled output. My friend's player has the 2134 with standard power supply (the PSR has to reject a load of digital hash) and 470uF output caps. After a lot of listening to a range of CDs using my Sonys and his HD600s I think that the direct coupled 4556 gave a slightly crisper sound than the 2134. The difference was marginal and it would be impossible to identify which player I was listening to in isolation. It was like when you go to the optomerist and you look at colored circles he asks "which lens gives the better vision" and you are thinking "well they are equally blurred". So my conclusion is that even under heavily biased conditions the distortions of both amplifiers are well below the distortions of the headphones and my ears.
ReplyDeleteThe other effect I referred to was the substitution of buffer amps which drive the cables to my speaker amps. Sorry I did not explain that well in my third comment. I still think that the 4556 sounds clearer than the 2134 (and 2604) when driving cables. 5m of cable seems to be a more difficult load than headphones.
I'll leave it there and not take any more of your valuable time. Thanks for pointing me to the 4556.
Steve
The schematic and parts values seemed vaguely familiar. I think the amp you bought on ebay is not a cmoy, but rather an RA-1 clone.
ReplyDeletehttp://diyaudioprojects.blogspot.com/2007/08/grado-ra1-headphone-amplifier.html
Except .... the feedback resistance is wrong, it should be 470k but ended up 470 ohms, so the gain drops from 6x to unity. Ebay. Shouda' known, right?