Pages

June 29, 2011

QRV09 DIY Headphone Amp

arv09-dscopeINTRO: A gentlemen in Sweden, Per-Anders Sjöström, has an interesting audio website and number of high quality diyAudio designs. I think he strikes a useful balance between solid engineering and audiophile excess. One of his most recent designs is the QRV09 Ultra High Performance Headphone Amplifier. The design generated a lot of interest at diyAudio and resulted in a QRV09 group buy of PC boards and various parts. Per-Anders has been very generous with his time considering he’s not likely to make any profit from the QRV09. I really admire “DIY Audio Philanthropists” sharing their knowledge and designs. But the one thing missing, from my perspective, was a proper instrumented test of the QRV09. So I decided to contribute some of my own time and try to fill that void.

THE DESIGN: The QRV09 uses very high-end Analog Devices AD8610 op amps for the input gain stage and the Texas Instruments TPA6120 as the output buffer. The TPA6120 is also used in the FiiO E9 I recently tested. The QRV09 has a relatively low gain of 2X (6 dB) but that’s easily changed. It includes an on-board AC power supply and jumpers to change the output impedance from 10 ohms to 120 ohms. You can read more on the Sjöström Audio Website.

NOT INCLUDED: The QRV09 is somewhat unusual in that it lacks an on-board volume control, power switch, and input/output jacks. So you’re on your own to implement those. The good news is that allows more flexibility configuring the QRV09 into a wider variety of enclosures and applications. It also allows using connectors and a pot of your choice rather than being limited by the PCB footprint. But its more difficult to use the board “standalone” by itself. I added temporary RCA input jacks, an IEC power receptacle, and a 1/4” headphone jack on a short “pigtail” for testing.

smt resistor pencilSURFACE MOUNT DESIGN: This project isn’t for beginners. You need experience working with surface mount technology (SMT) parts—specifically 0805 components which are only 0.08 by 0.05 inches or about 2 mm x 1.2 mm. The entire part is smaller than the tip of most pencils and there are about 60 of them on the QRV09 board. Three 0805 resistors in their paper tape are shown in the photo to the right with a normal pencil. Most find they need a special soldering iron tip, very fine gauge solder, really good tweezers, a magnifying lamp, flux, a very steady hand, and lots of patience. The other problem with SMT is once a part, even a resistor, is fully soldered it’s very difficult to remove or reposition without specialized equipment or unusual techniques. It’s a bit like gluing a model ship together inside a bottle. As one builder said on diyAudio, he was ready for brain surgery after assembling his QRV09.

TPA6120 SOLDERING: The TPA6120 output IC is a relatively small surface mount part that can, in theory, provide over 1 watt per channel of output but this requires it to dissipate nearly 3 watts of power. That means it needs a decent amount of PC board copper as a heatsink. TI, of course, thought of this and there’s a large metal thermal pad on the underside of the package. You can find various suggestions for how to solder this inaccessible pad but my personal opinion is it’s relatively difficult for a DIY’er to do correctly. And regardless of your technique, it’s even more difficult to verify you got it right. Per-Anders was kind enough to offer PCBs with the TPA6120 pre-soldered. My suggestion would be a large plated through hole big enough to get a soldering iron tip and some fine gauge solder through under the part. Then it would be possible to wick solder in from the bottom of the board and ultimately fill in the hole. The QRV09 board, however, has only tiny feed-through holes under the heatsink pad. TI assumed the part would be soldered with a commercial IR reflow oven and solder paste but these are not an option for most DIY builders. The TI datasheet notes if the pad is not correctly soldered to a suitable heatsink (copper PCB area) the IC may fail. See the Tech Section for more details.

USABILITY: If you don’t want to leave the QRV09 powered on 24/7, it does create a bit of a “thump” if your headphones are connected. How loud the transient is will depend on your headphones.  Also, I was surprised the Pulse Electronics transformers from DigiKey (553-1630-5) emit a soft buzz (from the transformer itself, not the headphones) that’s audible from a few feet away with the PCB exposed in a very quiet room. They’re the only readily available 120 volt transformers I know of in the USA that fit the PC board. Per-Anders used 240 volt European transformers that don’t have this problem. The QRV09 uses about $5 per year of electricity (USA rates) if left on all the time.

GAIN: The gain of the standard design is lower than normal at 2X (6 dB). This means to get the 2 volts RMS needed to drive say a pair of Sennheiser HD650 headphones, you need 1 volt RMS of input. That’s usually not a problem with a home pre-amp, home DAC, etc. But it might be a problem for a lot of portable devices and even some PC audio outputs. A gain of 3X or 4X is better if you want to use a wider variety of sources such as an iPod. I ran tests with both 2X and 4X gain. Many home sources are limited to about 2 volts of output and 2X gain limits the amp to 4 volts when it’s otherwise capable of over 7 volts. Considering how difficult it is to replace surface mount resistors, jumpers for selectable gain might have been useful. So choose your gain carefully before you build. The downside of too much gain is more noise, a less useful volume control range, increased volume tracking error and possibly more distortion. The FiiO E9 has a gain switch on the back to choose between 10 dB (3X) and 18 dB (8X) gain. See my article on amps and DACs for more.

qrv09 boardSOUND QUALITY: The stock design sounded fine with my higher impedance headphones. I mainly listened with my Sennheiser HD650s and didn’t notice any problems beyond the power on/off glitches mentioned above. It has more than enough power and, in casual comparisons, sounded a lot like my $1600 Benchmark DAC1 Pre’s headphone output. With my Ultimate Ears Super Fi 5 Pros, however, it sounded notably different than the Benchmark due to the QRV09’s 10 ohm output impedance. With a modification to lower output impedance it again sounded very similar to the Benchmark (see Tech Section for details).

HISS & NOISE: Even with my very sensitive Ultimate Ears the QRV09 was quiet but, as explained in the Tech Section, that’s at least partly due to the lack of a volume control and the abnormally low gain.

MEASUREMENT SUMMARY: The QRV09 measured very similarly, but generally a bit better, than the FiiO E9 which isn’t surprising as the FiiO E9 uses the same output IC. The E9 has a higher current power supply and delivered more power into 15 ohms but even the QRV09 delivers more power than anyone is likely to need at any impedance. The QRV09 is quieter and has lower distortion into higher impedance loads. They both suffer from the same 10 ohm output impedance which is far from ideal but OK for most higher impedance headphones (see my Impedance Article for why and the Tech Section for a possible fix). And they both have higher than expected distortion at low frequencies into low impedances. As usual, all the geeky details are in the Tech Section but here’s a summary and comparison

Measurement QRV09 FiiO E9 AMB Mini3
Frequency Response +/- 0.1 dB Excellent +/- 0.1 dB Excellent +/- 0.1 dB Excellent
THD 1 Khz 150 Ohms 0.002% Excellent 0.005% Excellent 0.002% Excellent
THD 1 Khz 15 Ohms 0.022% Good 0.037% Good 0.017% Good
THD 20 hz 15 Ohms 0.07% Good 0.05% Good 0.01% Very Good
THD 20 Khz 15 Ohms 0.02% Very Good 0.003% Excellent 0.45% Poor
IMD CCIF 0.02% Good 0.05% Good 0.043% Fair
IMD SMPTE 0.0015% Excellent 0.002% Excellent 0.009% Very Good
Noise (ref 400 mV) -99 dB Excellent (1) -88 dB Fair -94 dB Excellent
Max Output 15 Ohms 450 mW Excellent 1067 mW Excellent 104 mW Excellent
Max Output 150 Ohms 345 mW Excellent 317 mW Excellent 38 mW Fair
Output Impedance 10 Ohms Fair (2) 10 Ohms Fair 0.9 Ohms Very Good
Crosstalk 15 Ohms 67 dB Excellent 63 dB Very Good 40 dB Poor
Channel Balance Error N/A (3) 1.8 dB Fair 1.14 dB Fair
  1. The noise measurement isn’t directly comparable, see the Tech Section
  2. A modification is discussed in the Tech Section reducing this to 0.08 ohms which is excellent
  3. There is no volume control to cause tracking error

FIRST CLASS:

  • Excellent performance into higher impedance loads (greater than 64 ohms)
  • Lots of output voltage for high impedance headphones
  • On board power supply is convenient
  • Jumper for selectable output impedance (10 or 120 ohms)
  • Protected from RF and DC at the input

ECONOMY:

  • 10 ohm output impedance is a problem with many headphones
  • Disappointing distortion at low frequencies into low impedance loads
  • Gain is too low for many headphones using a portable or PC source
  • Does not include volume control, jacks or power switch
  • Requires considerable surface mount soldering skills to build
  • Inaccessible output IC thermal pad difficult to solder correctly
  • Uses some expensive parts (most notably the AD8610 op amps)
  • No headphone protection and has some “thumps” on power on/off

BOTTOM LINE: Unlike say the AMB Mini3, the QRV09 is a very solid design on paper. Per-Anders did an excellent job and the QRV09 follows solid engineering practice rather than attempting to solve problems that are more audiophile myth than reality. It also includes useful features the Mini3 lacks like RF and DC input protection. But, in my opinion, the QRV09 is another example of a design that looks great on paper but does not perform as well as expected in reality. My measurements uncovered some concerns that may disqualify the QRV09 as “Ultra High Performance”. But it still is very good into higher impedance loads. The performance is very similar to the FiiO E9. It’s a DIY project most can be reasonably proud of. But if you’re looking for the most bang for your buck, the FiiO E9 is probably cheaper when you consider all the costs. The E9 also adds several features including adjustable gain, line out, and a protection relay. In terms of ultimate performance, the Benchmark DAC1’s headphone amp is considerably better than either one. And just the headphone amp section of the DAC1 could be built for around the same price as the QRV09. But, again, things like PCB layout, grounding, etc. are critical if you hope to match the Benchmark’s performance. Ultimately, there’s no substitute for making the right measurements.


TECH SECTION:

DESIGN: The gain stage op-amps are individual left and right Analog Devices AD8610s that cost nearly $10 each at DigiKey. The TI TPA6120  is a somewhat unusual stereo current feedback power op amp with gain (it’s not a pure buffer like the National LME49600 or TI BUF634). Here it’s used at unity gain with only local feedback. There is no global feedback loop. The main downside to the TPA6120 is TI recommends at least 10 ohms of series output resistance to isolate reactive loads for stability reasons. As I’ve explained many times, including in my Impedance Article, I think that’s unacceptably high for many headphones—especially balanced armature IEMs such as those from Ultimate Ears, Etymotic, Shure, Westone, etc.  But it will also affect the sound of many full size dynamic cans.

INPUT FILTER: The QRV09 intelligently includes an RF filter on the input. In this age of ubiquitous mobile phones it makes good sense. RF is easily rectified to DC at the input of an op amp and that can create substantial DC on the output causing unpleasant noises, and worse, damage expensive headphones. The QRV09 uses 1K in series and 100 pF to ground to help kill RF. It also wisely includes a coupling capacitor to block DC from the source. Per-Anders used 0.1uf Polyphenylene Sulphide (PPS) 2416 SMT film capacitors. The amp I tested used 0.12 uF Panasonic ECH series PPS caps. There’s also a “bleed” resistor on the input side of the capacitor eliminating any residual DC. The 0.1 uF cap and 470K resistor form a high pass filter that blocks DC with negligible effect in the audio range. Good film coupling caps have been demonstrated to be sonically transparent in blind tests, measurements, and audio differencing. So I very much agree with Per-Anders’ design decisions addressing real world problems of stray RF and DC offset at the input without any sonic sacrifice.

NICE LITTLE TOUCHES: The power supply has extensive filtering capacitors and even uses a MOV and fuse for protection. There are board-to-wire connectors for all inputs and outputs. Per-Anders clearly sweats the details.

GAIN MODIFICATION: I ran most tests with the “stock” 2X gain. I also spot checked several applicable measurements after changing the gain to about 4X by paralleling a 470 ohm resistor on top of R4 and R16 (1K to ground on the negative op amp input).The parallel result is 320 ohms which gives a gain of 1 + 1000/320 or 4.1X (12.2 dB). Most of the graphs indicate 2X or 4X. As expected the noise is about 6 dB worse at the higher gain setting. The ultimate crosstalk into 100K was also reduced by several dB. But the distortion measurements were mostly very similar indicating the input op amp is not the dominant source of distortion.

qrv09 ferritesOUTPUT IMPEDANCE MODIFICATION: I tried several ways to reduce the output impedance. I ultimately found some specific 0805 chip ferrites that seem to provide sufficient reactive load isolation and stability while providing a near zero ohm output impedance. The ferrites are from Laird and the part number is HI0805O121R-10 (DigiKey 240-2393-1). These are rated at 120 ohms at 100 Mhz and 3.5 amps DC with a 20 milliohm DC resistance. I tried several ferrites and several inductors including 0805 chip inductors, conventional through hole inductors, and an air core inductor. The ferrites provided the best square wave performance, and perhaps more important, offer easily reproducible results. Any “off board” solution would be more prone to implementation details, and where stability is concerned, that can make a big difference. The ferrites are simply paralleled and soldered on top of the 10 ohm output resistors R10 and R22 as shown in the photo to the right (click for a larger version). The 10 ohm resistor remains to provide damping for the ferrite’s resonance. There are, however, some trade offs to reducing the output impedance this way. I discuss them in the measurements below.

THERMAL ISSUES AND A SOLDERING CHALLENGE: As mentioned earlier, the TPA6120 heatsink pad is very difficult to properly solder. I played typical pop music, hitting peaks of only 1 volt, into a 15 ohm load for 15 minutes and the TPA6120 plastic package was at 120 F or about 50C. The die inside was probably over 70 C. That’s not dangerously hot, but it’s a possible sign the thermal pad is not working as well as TI intended due to poor soldering. Driving 300 ohm headphones to typical levels the pad is probably a non-issue. But in more demanding applications, or with high ambient temps, it might be. The other danger is someone could easily damage the part, perhaps in non-obvious ways, getting it too hot trying to solder the pad. One possible solution is not even trying to solder the pad and use lower supply rails (which can be done with just resistor value changes on the regulators). That would substantially decrease the dissipation but also the maximum output swing into power hungry high impedance cans. In theory, the part is thermally protected but quoting the datasheet regarding the hidden pad:

This acts as a heatsink and must be connected to a thermally dissipating plane for proper power dissipation. Failure to do so may result in exceeding the maximum junction temperature that could permanently damage the device. “

FREQUENCY RESPONSE: There’s a slight roll off of about 0.3 dB at 10 hz from the DC blocking input cap but the response is  +/- 0.1 dB from 20 hz to 96 Khz. The –3 dB point, even with the ferrites in place, was 220 Khz. Some might argue this is excessive bandwidth but it does keep the phase error low. The channel balance, with no volume control, was excellent (within 0.004 dB). The frequency response below 48 Khz was independent of the load impedance:

QRV09 2X Frequency Response 150 Ohms (ref 400 mV RMS)

THD vs POWER OUTPUT STOCK: The power output in stock (10 ohm) form was ample but significantly less into low impedance loads than the FiiO E9 using the same TPA6120. With some investigation I found the power supply is falling out of regulation (see Power Supply Constraints below). Still, 450 mW into 15 ohms should be far more than anyone needs and far in excess of what the Mini3 and most headphone outputs can manage. And the dynamic (short term peak) power is much higher (closer to the FiiO E9) due to the 470 uF caps after the regulators. THD+N into 150 and 600 ohms was very low and noise dominated the measurement below 1 volt. The distortion was quite a bit higher into low impedances but still generally below about 0.01%. The graph shows output voltage vs THD+N at 1 Khz into four load impedances. The power is shown at approximately 1% THD for each load:QRV09 2X THD N vs Output 15 33 150 & 600 Ohms (top to bottom) comments

THD vs POWER FERRITE MOD: With the ferrite modification you get somewhat more power into 15 ohms as you’re no longer dropping nearly half the power across the 10 ohm output resistors but the limiting factor is still the power supply. It’s also worth noting the ferrites roughly double the distortion from about 0.012 to about 0.025% across much of the range:

QRV09 2X THD N vs Output 15 Ohms Ferrite (blue) No Ferrite (green) comments

POWER SUPPLY CONSTRAINTS: The 12 volt rails drop out of regulation at about 1 volt RMS into 15 ohms with both channels driven by a sine wave. They’re down to 8 volts at 2.5 volts RMS into 15 ohms. The SOT223 regulators are rated for 1 amp, which is plenty, but the the small AC transformers can’t handle the load. Some might be bothered by the sag but it’s not likely to happen in real world use with music, typical loads and realistic output levels. And there is even some benefit if the rails do sag. The dissipation in output devices increases exponentially with the supply rail voltage. Amps are more efficient, and produce less heat, when the rail voltage is matched to the load. That’s why some power amps have impedance switches on them and others use multi-rail or “smart” tracking power supplies (i.e. Class G, H, etc.). Hence the QRV09’s “soft” power supply helps reduce power dissipation in the TPA6120 under extreme use when the TPA6120 is most likely to get too hot. Soft power suppliers are common in mainstream amps and receivers as they allow manufactures to use smaller heatsinks and save money. The small transformers also fit well on a reasonably sized PCB and into slim enclosures.

OUTPUT IMPEDANCE: With the 10 ohm resistors in place in the stock design, 400 mV no load dropped to 240 mV into 15 ohms giving, as expected, a 10 ohm output impedance. With the jumpers removed the output impedance was 120 ohms. And with the ferrite modification 400 mV no load barely dropped to 398 mV with 15 ohms giving an output resistance of 0.08 ohms—the lowest I’ve yet measured. Even at 20 Khz it only rose slightly to 0.09 ohms. This is excellent performance with the ferrite mod. In case you think 10 ohms output impedance is just fine, here’s what 10 ohms does to the frequency response of my Ultimate Ears SuperFi 5 Pro headphones (from the FiiO E9 which has the same TPA6120 and same 10 ohm output resistor):

FiiO E9 Frequency Response Large Jack Ultimate Ears Super Fi 5 Pro

THD+N 1 KHZ 150 OHMS: Here’s the impressively low distortion at 1 Khz and 1 volt into 150 ohms. For comparison, the FiiO E9 measures about 0.005% on this test. Also note the channels are well matched which indicates a symmetrical PCB layout:

QRV09 2X 1 Khz Wideband Spectrum & THD N 150 Ohms (Ref 1V RMS)

THD+N 1 KHZ 15 OHMS: There’s about 10 times the distortion into 15 ohms with the ferrite. Without the ferrite, this drops to about 0.012% but that’s still about 5 times the distortion compared to 150 ohms. That said, this is still a significantly better result than the FiiO E9 which was nearly 0.04%. And it’s mainly 2nd harmonic which is the most benign. This is likely inaudible as the higher harmonics are all below –90 dB:

QRV09 2X 1 Khz THD N Wideband 15 Ohms (ref 400 mV RMS)

THD RESIDUAL 150 OHMS: Here’s the 1 Khz residual at 150 ohms and there’s perhaps a hint of crossover distortion but it’s fairly benign:

QRV09 2X 1 khz Residual 150 Ohms (ref 400 mV RMS)

THD RESIDUAL 15 OHMS: Here’s the 1 Khz residual at 15 ohms showing it’s mainly 2nd harmonic with no signs of crossover distortion (note the different vertical blue scale compared to above):

QRV09 2X 1 khz Residual 15 Ohms (ref 400 mV RMS)

THD 20 KHZ 150 OHMS: Into 150 ohms the high frequency distortion, measured out to 88 Khz, was very low:

QRV09 2X 20 Khz Wideband Spectrum & THD N 150 Ohms (Ref 1V RMS)

THD 20 KHZ 15 OHMS: Into 15 ohms, with the ferrite, it’s notably worse but still not much to worry about with no spectral components above –80 dB anywhere in the audio range. See the Frequency vs THD graph for the distortion without the ferrite. Here the FiiO E9 has the advantage. It measured 0.003% on this test. I suspect the FiiO has a better PCB layout for routing the higher currents required at 15 ohms and/or uses different feedback:

QRV09 2X Ferrite 20 Khz THD N 15 Ohms (ref 400 mV rms)

THD 20 hz 150 OHMS: Into 150 ohms the QRV09 does very well at 20 hz with everything below –90 dB and the 2nd harmonic dominant. (Note the THD reading is slightly higher than the THD+N reading because it uses a different notch filter to remove the fundamental while leaving only the harmonics. Due to the “excessive” width of the fundamental at 20 hz, even with a 256K FFT, more of the fundamental leaks into the result. I’ll remove it from future 20 hz results):

QRV09 2X 20 hz Spectrum & THD N 150 Ohms (Ref 1V RMS)

THD 20 hz 15 OHMS: Into 15 ohms (no ferrite), just like with the FiiO E9, things get worse. The good news is the 2nd harmonic still dominates and is the only non-linearity over –80 dB but it’s a lot over at about –62 dB. Everything else is below –80 dB. The ear is less sensitive to low frequency distortion and the 2nd harmonic, so this is likely still inaudible, but this isn’t what I would call “ultra high performance”. The FiiO E9  measured a similar 0.05% on this test. If you’re thinking the dScope is somehow to blame, have a look at the Cmoy Review to see 0.0048% into 15 ohms—about 10 times lower--and the Benchmark DAC1 measures much better as well:

QRV09 2X 20 hz THD N vs Frequency 15 Ohm (Ref 400 mV RMS)

THD 20 hz 15 OHMS 100 mV: One source of low frequency distortion can be thermal distortion. Wondering if that’s to blame here, I tried reducing the level to only 100 mV and driving only 1 channel but it didn’t help much and the 2nd harmonic is still about –65 dB.

QRV09 2X 20 hz THD N 15 Ohms One Channel Driven (ref 100 mV RMS)

THD 20 hz 150 OHMS WITH FERRITE: Here’s the same test as above with the ferrite. The ferrite roughly doubles the distortion and much of the increase is in higher order harmonics. This is where an air core inductor has an advantage but, in my tests, created significantly more ringing on the output and is more subject to implementation issues. Regardless, based on the testing I’ve done, the TPA6120 forces you to trade off distortion, output impedance and ringing/impulse response. The best solution will depend on what headphones you’re using. With balanced armature IEMs a 10 ohm output impedance is unacceptable. So it comes down to more ringing vs a bit more distortion. For the sake of maximum stability and repeatability, I chose the higher distortion. At 80hz it’s only slightly above –80 dB so it’s still very likely inaudible. To me, it’s the best compromise but an even better solution is to use a better output device than the TPA6120 that’s stable without anything in the output (ignore the typo of “10 hz” in the caption—you can see it’s 20hz in the spectrum).:

QRV09 2X Ferrite 10 hz THD N 15 Ohms (ref 400 mV rms)

THD+N vs FREQUENCY: Here’s the THD vs frequency into both 150 ohms (yellow) and 15 ohms (blue) and it looks amazingly like the FiiO E9 graph shown below. There’s a steady rise in distortion into low impedance loads below about 5 Khz. This is the stock 10 ohm design without the ferrite. I don’t know if this is a PCB layout issue or some inherent problem in the TPA6120 but it’s not very impressive. The Cmoy I tested, with it’s $0.80 op amp, does much better at 15 ohms:

QRV09 2X THD N vs Frequency 150 Ohms (yellow) 15 Ohms (blue) (Ref 400 mV RMS)

FiiO E9 FOR COMPARISON: Note how similar the FiiO E9 curves are below to the QRV09 above:

FiiO E9 THD vs Frequency Yellow = 150 Ohms Blue = 15 Ohms (400 mV to 22 Khz)

THD+N vs FREQUENCY WITH FERRITE 150 OHMS: Here’s the 150 ohm plot with (blue) and without (yellow from above) the ferrite. Only below about 200 hz does the ferrite significantly add any distortion and it’s still nothing to worry about:

QRV09 2X THD N vs Frequency 150 Ohms Ferrite (blue) (ref 400 mV RMS)

THD+N vs FREQUENCY WITH FERRITE 15 OHMS: Here’s the comparison with and without the Ferrite at 15 ohms and due to the much higher current it, as would be expected, causes more distortion. You can see some weirdness above about 5 Khz and overall higher distortion:

QRV09 2X THD N vs Frequency 15 Ohms Ferrite (blue) (ref 400 mV RMS)

IMD CCIF: Here’s the CCIF IMD at 15 ohms. This is significantly better than the FiiO E9 which measured 0.05%. The –72 dB component at 1 Khz is a 2nd order distortion product from the difference of the 19 Khz and 20 Khz signals. Ideally, I’d like to see it below –80 dB. The good news is the higher order sidebands around the 19 and 20 Khz tones are very low at about –105 dB:

QRV09 2X CCIF IMD 15 Ohms (ref 400 mV RMS)

SMPTE IMD 150 OHMS: Into 150 ohms the SMPTE IMD was extremely low (just about as-good-as-it-gets—the IMD distortion is mostly lost in the noise and entirely below –110 dB):

QRV09 2X SMPTE IMD 150 Ohms (ref 400 mV RMS)

SMPTE IMD 15 OHMS: Into 15 ohms there’s mostly just the higher harmonic (not IMD) distortion from the 60 hz component. The actual IMD is still very low at 0.0015% which is slightly better than the FiiO E9’s 0.002%:QRV09 2X SMPTE IMD 15 Ohms (ref 400 mV RMS)

INTERCHANNEL IMD: I ran some interchannel tests and the results were excellent as would be expected with isolated power supplies and a real ground. This review is already quite long so the graphs are not shown. The result was almost entirely dominated by the crosstalk from the headphone jack.

CHANNEL SEPARATION (CROSSTALK): I should have explained in previous reviews the crosstalk in a headphone amp is sometimes limited driving real world headphones by the 3 wire connection. Both channels share a common connection, and the lower the load impedance, the more that connection contributes to the crosstalk. Most other devices I’ve tested have had factors dominante. It comes down to the PC board layout, the resistance in the jack itself, any wiring, grounding (or, worse, virtual grounding), how the feedback is configured, etc. I minimize the resistance of any external wiring when I attach the load which simulates using headphones with 4 wire cables. Still, even the jack resistance typically limits crosstalk to around 70 - 80 dB into 15 ohms and around 80 - 90 dB into 150 ohms. The QRV09 was indeed in this range so I also measured it at 100K to show (somewhat unrealistic) crosstalk within the amplifier itself. The results are shown below into 15 ohms (blue), 150 ohms (yellow) and 100K ohms (red). All three are excellent performance and better than the FiiO E9:

QRV09 2X Crosstalk 15 150 100K Ohms Top to Bottom (ref 400 mV RMS)

NOISE: Reproducing a –80 dB signal, the noise performance was the best I’ve measured but there are two reasons why it’s not a fair comparison. First, there’s no volume control and the dScope has a very low source impedance minimizing the Johnson noise. The AD8610 is so quiet the broadband noise is nearly entirely the Johnson noise of the 1K input resistor. With a typical 10K volume control in place there would be at least twice as much Johnson noise so the noise floor would increase by at least 6 db. Second, this amp has only 2X ( 6 dB) gain. Increasing the gain to a more typical 4X (12 dB) increases the noise floor another 6 dB. Note this is referenced to the same realistic 400 mV I always use. Referenced to full output (7 V RMS), the way a manufacture might spec it, the noise is about 20 dB better (-118 dB unweighted –126 dB weighted). The nearby AC transformers from the onboard power supply appear to be the limiting factor. Somewhat odd is the amplitude of the 180 hz artifact of the 60hz AC power and 120hz ripple. But everything is still below –110 dB which should be silent. The actual hiss portion of the noise (above 500 hz) is about 16 dB quieter than the FiiO E9:

QRV09 2X Noise 150 Ohms (ref 400 mV RMS)

PHASE RESPONSE: The phase response below 48 Khz was unchanged by the ferrites and the result below is with the ferrites. The input DC blocking capacitor causes the expected shift but it’s still only about 8 degrees at 20 hz. At 10 Khz the shift is an excellent 0.25 degrees. But this is to be expected with a bandwidth beyond 200 Khz:

QRV09 2X Ferrite Phase Response

GAIN: The gain measured 2X or 6 dB in the stock configuration and and 4.1X (12.2 dB) with my modification. The higher gain slightly increased the 100K crosstalk by several dB and increased the noise by 6 dB. It had very little effect on distortion which speaks well for the AD op amp’s gain bandwidth product and distortion but also indicates the TPA6120 and/or other factors are probably the weak link in terms of distortion. Other gains are, of course, possible just by changing the feedback resistors on the op amps.

CLIPPING BEHAVIOR: Input clipping was not an issue. The output will always clip first for any gain greater than 1X. Even into reactive loads the amp was stable at clipping with and without the ferrite modification. Here’s clipping with the ferrite into 600 ohms in parallel with 0.01 uF at 1 Khz. The TPA6120 gets fairly close to the supply rails and the clipping is nicely symmetrical:

QRV09 4X Inductor 1 Khz Clipping Scope Trace

CLIPPING SPECTRUM: Worried about possible instability I checked the spectrum while clipping and there was no unexpected ultrasonic garbage out to 96 Khz with the ferrites. I didn’t try sustained clipping into 15 ohms for fear of damaging the TPA6120 if the thermal pad isn’t properly soldered:QRV09 4X Inductor 1 Khz Clipping Wideband Spectrum 150 Ohms (rev 1 V RMS)SQUARE WAVE PERFORMANCE STOCK: With the same Sennheiser CX300 headphones I always use as a load, the stock QRV09 exhibited significant overshoot on a 10 Khz square wave. I’m guessing something is not quite right here because the FiiO E9 did not have this issue. It might be related to the PCB layout or perhaps input/feedback resistor values used for the TPA6120:

qrv09-4x-10khz-squarewave-sennheiser-cx300

SQUARE WAVE PERFORMANCE WITH FERRITES: With the same CX300 headphones as above, but the 10 ohm output resistors paralleled with chip ferrites, the overshoot is gone. The red trace is just the headphones and the blue trace is the headphones with a 0.047 uF film capacitor in parallel. You can see a small amount of 500 Khz ringing related to the added capacitance, and a bit slower rise time in both traces, but the amp is stable:

qrv09-10khz-squarewave-47nf-blue-sennheiser cx300 ferrite

SLEW RATE WITH FERRITE: The slew rate shown below between points 1 and 2 is 11.27 volts in 688 nS. That’s about 16 volts per microsecond. This is very fast and about double the FiiO E9. But, in fact, both amps need less than 2 volts/uS. The main point here is the ferrite chip doesn’t come close to causing speed problems even at full swing of over 20 V p-p into 600 ohms:

qrv09-slew-rate-with-ferrite

STABILITY WITH FERRITES: I conducted a lot of tests using various ways to bypass the 10 ohm output resistors while still isolating reactive loads for stability. I tried 1 ohm chip resistors and, with added capacitance on the output, observed signs of significant instability. A 1 uH chip inductor created more ringing and more distortion in the audio band compared to ferrites. And an air core “off board” inductor also created more ringing but less distortion. The classic DIY solution for audio power amps intended to drive speakers is a wire wound coil around a 1 or 2 watt carbon composition resistor. That created more ringing than the chip ferrites. As a torture test, I used about 20 feet of headphone extension cables strung together with a “Y” cable at the end. I used up to 1.5 uF of capacitance by itself and in parallel with various real headphones and resistive loads. I also tried lots of different headphones by themselves. Into all the loads, I ran a low level 10 Khz square wave and observed the response on a 100 MHz scope, as shown above, for signs of ringing or oscillation. Overall, I was most satisfied with the chip ferrites especially when you consider their implementation is nearly foolproof (unlike handwound coils and other off board solutions). Another thing to try would be to replace (or parallel) the 10 ohm resistors with 1 ohm resistors and add through hole ferrites to the 120 ohm jumper locations.

DC OFFSET: The DC offset measured an excellent 1.8 mV in one channel and 2.3 mV in the other. The expensive op amps likely help here.

POWER CONSUMPTION: The design draws about 40 mA AC at 120 volts or 4.8 VA at idle. The bulk of this are losses in the transformers. If left on 24/7 that’s about $5 per year in electricity. The DC power consumption is also about 40 mA from the 12 volt rails or only about 1 VA.

DESIGN COMMENTS: For anyone considering the QRV09. or a similar design. here are some things to consider:

  • Output Impedance - As with the FiiO E9, and likely most other TPA6120 headphone amps, the 10 ohm output impedance is a problem for many headphones. I explain why in my Impedance Article and an earlier graph shows big (6 dB) frequency response problems into my IEMs. It also degrades electrical damping and hence the bass performance of many full size headphones like the Denon AH-D2000s. In my opinion, it prevents the TPA6120 from being a “one size fits all” high quality amp. Reducing the output impedance to near zero using on-board ferrite chips increases distortion somewhat (in many cases roughly doubling it). Some further experimentation might yield a better compromise and a more ideal output network. The danger here is there may be some headphones that could trigger instability and potentially damaging oscillation in the TPA6120 when the 10 ohm resistors are bypassed. So unless you have the means to fully test the stability of your particular project, I suggest care in trying alternate techniques. I chose the chip ferrites because they’re easily reproduced by others, relatively fool proof, and seemed like the best overall compromise (see Stability With Ferrites above).
  • Low Frequency Distortion - The rising distortion at lower frequencies into lower impedance loads is a puzzle. When I observed it with the FiiO E9 I assumed it might be related to the DC-DC power supply in the E9. But it’s nearly exactly the same here with the QRV09 using a very conventional power supply. The problem is either inherent to the TPA6120 or both designs somehow have a very similar design flaw. One possibility is how the power supply and ground currents are handled. The QRV09 layout uses ground fills on both sides of the PCB rather than specific ground routing. That might cause higher distortion at low impedances but I would expect it to be similar at all frequencies. Anyone have any ideas as to the cause of low frequency distortion? 
  • Dual Mono Questions - I’m not sure there’s much benefit to using two single op amps and two sets of power supply regulators when the TPA6120 is a single stereo IC and the separation is ultimately limited by the 3 wire output. You could dedicate a TPA6120 for each channel (and even parallel them) but in terms of the weakest link, the headphone jack is, by far, the biggest offender. That said the QRV09 is slightly better than the FiiO E9 in a few areas and the dual mono configuration might be partly responsible.
  • TPA6120 – I’m not sure the TPA6120 is the best choice. I can understand why FiiO and other commercial manufactures use it as it’s relatively inexpensive (less than $2 per channel in volume). Their products are also manufactured with commercial IR reflow ovens and screened solder paste eliminating the soldering problems. But given the difficulty in properly soldering the heatsink pad by hand, the 10 ohm output resistor requirement, and the mysterious low frequency distortion, I consider the TPA6120 a marginal choice for a high quality DIY amp. On paper it looks good. But, for the two TPA6120 designs I’ve measured, the Benchmark DAC1’s headphone amp easily outperforms both and even the Cmoy I tested outperformed it in many ways into 15 ohms.
  • Op Amp Overkill - I’m not sure the AD8610 is providing any meaningful benefit over several less expensive op amps. The real world noise is limited by the Johnson noise of the input circuitry and hum components from the onboard power supply. And the distortion is apparently more limited by the TPA6120 and/or other factors. The much cheaper OPA2134 in the FiiO E9 delivers very similar performance but it’s a dual op amp and not pin compatible. But the 2134’s single channel cousin, the OPA134, and the National LME49710, would work nicely. Both are about 1/4 the price of the AD8610. I have a future article planned on “op amp rolling” to explore any differences more.
  • Output Relay – The FiiO E9 has an output relay with delayed turn on and immediate disconnect on power off. Given the QRV09’s audible “thumps” on power up and down I understand why FiiO went to the added expense. An output relay can also be used as DC protection to help save expensive headphones if the amplifier fails.
  • Gain Selection – 2X gain might not be enough for some applications. Think carefully about your gain requirements before selecting the values for R4 and R16. See the gain comments in the review.

TECH SECTION SUMMARY: Generally the QRV09 measures better or similarly to the FiiO E9 except in high frequency distortion and maximum power output into low impedances. The FiiO E9 is a very similar design using the same TI TPA6120 so this isn’t surprising. The FiiO has a higher current power supply while the QRV09 is quieter, has less crosstalk, and lower distortion into 150 ohms and above. The biggest issue for both of them is the output impedance. See my comments in the Design Comments immediately above. Given the output impedance issue, and the relatively high levels of low frequency distortion, some might not consider the QRV09 (or the E9) “ultra high performance” but the performance is still very good into higher impedances. And, unlike the AMB Mini3, the distortion into even low impedance loads is likely still inaudible for the reasons mentioned. But, in terms of raw performance, the Benchmark DAC1 delivers significantly better performance than either one of them. And the Bechmark’s headphone amp section could be built for around the same price. But, that said, the QRV09 and FiiO E9 both sound and perform just fine with a lot of high impedance headphones including my Sennheiser HD650s. The QRV09 might not have the ultimate specs but it’s a solid design.

13 comments:

  1. Another fine entry. I feel you re: small parts. I'm working on a USB audio dongle for a buddy that has developed a weird RF pickup issue. Christ - I'm too old and shaky to fiddle with such tiny parts...

    Rob

    ReplyDelete
  2. Another great article. To bad I'm way too clumsy to something like that together

    I like the idea of building a clone of the Benchmark's amp. I'd especially love it if there were no SMDs so I could do it myself.

    ReplyDelete
  3. When there are no revolts over product reviews, it looks like you get a lot less comments, heh.

    I'm looking forward to the entry on op amp swaps for sure. It always seemed dubious to me.

    If you're modifying a relatively high-end design, you're probably not going to get significantly better performance by using something else, and you might end up with something a lot worse. Somebody picked the stock op amp hopefully with a good reason.

    If you're modifying a relatively low-end design that uses a fairly low-end op amp, then maybe you could replace it with something significantly more exotic, with higher specs. But then which to use? The circuit wasn't designed with the different op amp in mind, and on a lower-end product, the overall design, other components, and layout are also likely to be limiting factors, not just the op amp.

    Some of the popular op amp alternatives seem to have overkill bandwidth and other issues as well. Regardless, I'm sure you could find some setup where upgrading the op amps does make a significant difference. It would just be the exception to the rule, I'd guess.

    ReplyDelete
  4. Mike, thanks for the comments. There are some examples where "upgrading" op amps does change the sound in audible ways--sometimes for the worse. I'll be covering that in the future article.

    For the QRV09 I think there are several less expensive op amps that would deliver sonically identical performance. So, in this case, it's not about better sound but saving money. Per-Anders was aiming for extremely high performance so I can understand his choice but some might want to save $15.

    ReplyDelete
  5. Oh whoops, I meant to write difference "for the better."

    About the 0.004 dB channel balance error on the QRV09: this is primarily the result of individual component tolerances on each side of the circuit? i.e. actual resistance values of parts being +- 1% or whatever of the nominal value.

    I didn't look at the circuit other than the photograph displaying how...symmetrical it was, but intuitively that channel balance error seems smaller than I would have expected with off-the-shelf parts, even if they're relatively expensive. I suppose it's all in the design.

    ReplyDelete
  6. Mike, the channel balance is indeed up to the resistor tolerances and the relative gain. But it's also part luck. Some resistors are machine graded into tolerance categories during manufacturing. If you're buying the tightest tolerance for that particular series of graded resistors some of the values will be almost perfect. But, if you buy one of the lesser tolerances, none of them will be close to perfect because those all ended up sorted into the better grade. In this case 1% is probably the tightest tolerance for that series of resistor and the values in the QRV09 worked out to match between the channels.

    The board is almost perfectly symmetrical with a couple minor exceptions. And, unlike some of the Asian design boards I've seen, Per-Anders seemed to optimize the layout for best performance rather than nice aesthetics.

    ReplyDelete
  7. "And, unlike some of the Asian design boards I've seen, Per-Anders seemed to optimize the layout for best performance rather than nice aesthetics."

    That sound like a premise for another article all by itself.

    ReplyDelete
  8. Another great in-depth article. I'm rapidly becoming addicted to your site and I've started playing with op-amps again which is fun. Something I would like to see is a large photo (under good lighting) showing each side of the full PCB for any equipment you mention. Thanks for all the work you put into organizing and presenting these investigations.

    ReplyDelete
  9. Regarding the increased distortion with the ferrite and 15 Ohm load, isn't it largely because the amplifier then needs to drive a more difficult load of 15 Ohms only instead of 25 total ?

    ReplyDelete
    Replies
    1. No, it's mainly the ferrite. I've seen the same issue with other amps, even without any series resistance involved. Plus the TPA6120 has very good low impedance drive and current capability.

      Delete
    2. Low frequency distortion with low impedance loads, even at relatively low output levels, is a common problem in TPA6120 based amplifiers, including the FiiO E9, the QRV09 measured here, and the Xonar Essence ST and STX sound cards. All of these are quite similar in this regard.

      A load of 15 ohms with the ferrite is similar to 5 ohms without the serial resistor being bypassed, so the increase in low frequency distortion is not surprising. However, the rise in distortion at the high frequency range (not typical of the TPA6120) is indeed likely caused by the ferrite, of which the impedance is also increasing with frequency.

      Delete
    3. The LF distortion is odd and a serious weakness of the TPA6120 (along with the 10 ohm series resistance requirement). I don't, however, agree a 15 ohm resistive load with the series ferrite is "similar to a 5 ohm load". At low frequencies the ferrite isn't adding significant reactance and, regardless, its effects are in series with the 15 ohm resistive load.

      Delete
  10. I meant similar to a 5 ohm load with the original 10 ohm serial resistor, so the TPA6120 sees a total of 15 ohms in the audio band in either case.

    ReplyDelete

PLEASE NOTE! If you see "LOAD MORE..." OR "LOADING..." just above this paragraph, you need to click that link to load the newest comments. It may take several seconds. When all the comments are displayed the Load More/Loading link disappears.

COMMENT APPROVAL: Comments may take anywhere from a few minutes to a few days to be approved. You might want to save a copy (highlight Ctrl-C) of your comments before submitting in case something goes wrong. The approval process is necessary to prevent obscenity, etc. Please keep comments on topic. Comments promoting commercial interests will not be approved and may be reported as SPAM.