Powering a high-end USB DAC

JanErik

Member
2010-08-09 9:35 am
Vasa
I am planning to build a high-end USB DAC with PCM2707 and PCM1794. 5V is of course available and 3.3V is easily done. But it will also need +-12V for output opamps.

Is there any sanity in powering this over USB and including an SMPS for the opamps?

Another possibility is to make it internal in the PC, and get 12 V from there, and then just use a voltage inverter circuit for the negative part.

I would really not want to include a mains transformer for this.
 
Last edited:

twest820

Member
2009-06-24 10:49 pm
Output swing (and thereby output impedance) is specified in the 49740 datasheet. You can also look at the output curves in the 49710 datasheet. Is that enough? Well, it depends on your system's gain budget and interconnect SnR requirements. Keep in mind -132dBFS DnR is 1uV of noise on a 4V swing. Like most Burr-Brown/TI DACs the PCM1794A is not specified for PSRR and does not have an eval board so the workback from 1uV to the required supply filtering is not exactly straightforward. For 5V USB in to 5V to the DAC I would start by looking at a CLCLC filter and seeing how low the cutoffs can be pulled---Cirrus uses about 4kHz on their eval boards but just because Cirrus parts can tolerate this doesn't mean TI's are similarly class A in the upper audio bandwidth. You might try TDK's MLZ series as the "dropout" would be around 30mV max---difficult to compete against with a regulator as LDOs like the TPS73101 or TPS7A4901 don't perform well at the low dropout corner.

Holding a 1uV noise floor across an interconnect probably implies a differential receiver with 0.1% resistors to get 60dB CMRR so that up to 1mV of ground bounce across the interconnect can be tolerated without throwing away too much of the DAC's DnR. So you might also look at the LME49724.

For -5V I'd consider something like an MAX1673 entry level for this application. Given the limited headroom here post regulation with a negative LDO like the TPS7A3001 isn't particularly attractive. A more desirable (and simpler and lower cost) solution is to hold the switching frequency far enough above the audio band an LC supply filter can provide good ripple rejection. If you can guarantee a sufficiently high load impedance the IV/output buffer operates class A the requirement relaxes as the LC cutoff can be lowered into the audio band.

If you go +-12 or do boost conversion things are easier as there's more headroom to work with. But I'd give +-5V and LC pi filtering a close look first.
 
Last edited:
You can power it however you want, the OP amps and DAC have their requirements, one thing I would try to break is the tie to the PC ground. If USB is used then find a modular DC DC converter with low parasitic capacitance between input and output. The high conversion frequency they use can be filtered with a compact LC filter, I would include a common mode inductor right at the output and before the first filter capacitor. Keep the corner frequency below 10 kHz if using a linear post regulator most do not have good rejection of higher frequencies. This will allow the use of any voltage even +/- 15V as long as the USB can supply the power.
As the previous poster mentioned the LDO regulators like a bit of headroom, read the respective data sheets to see the optimum value for line and load regulation.

Another vote for differential output, OTOH why not convert the USB signal to an optical signal and put the DAC in the power amplifier case? AFAIR USB requires 2 signal wires and optical transciever pairs are common as are multimode fibre pairs to suit. Anyway some food for thought, I set up instrumentation in noisy environments and always try to digitize the signal as close to the source as possible then send the signal out over an isolated link such as ethernet or better still ethernet over fibre.
 

twest820

Member
2009-06-24 10:49 pm
metalsculptor, do you happen to have measurements indicating the point at which isolation becomes necessary? For example, the CS4398 USB DAC I'm listening to whilst typing this uses CLCLC supply filtering and nonisolated grounds and measures at -100dB THD+N and 108 dBA DNR. This is a 4uV noise floor and just about best case datasheet THD peformance on the MAX4477 RRO op amps used in the output buffers. There are comparable RROs but the 4477's are the best overall tradeoff I'm aware of on the market so, given the 5V single supply and unbalanced output implementation choices the designers made for this headphone DAC, it's hard to argue with the parts choices or design quality. A differential output implementation should default to 114 dB DNR due to the swing doubling and I don't think it'd be hard to reduce the output noise by 1+uV by removing the explicit ground reference and switching to more capable op amps for the output buffer. That would result in 120+ dB DNR, which is pretty darn decent.
 
metalsculptor, do you happen to have measurements indicating the point at which isolation becomes necessary?

No that is the whole problem with ground noise, every application can be different, Your current set up may work fine but there is no guarantee that the next system modification will not change all that. While I am all for measurement, ground noise is so ad hoc and people here appear to spend lots of time either tracking it down or blaming it for something, why not just design it out and forget about it when the signal is digital there is little reason not to isolate, analogue isolation is more problematic. From my experience with industrial signals, isolated systems are so much less trouble than non isolated.
 

twest820

Member
2009-06-24 10:49 pm
Hmm, I'll differ somewhat. Ground behavior is deterministic---it obeys KCL and KVL, after all---but is a different kind of circuit from what most folks work with in that most of the impedances one cares about are parasitics and much of the loading is from things which are usually ignored. Low impedances mean low voltage levels and measuring things therefore requires care, which means acquiring the data to deterministically troubleshoot a problem is beyond the measurement tools most DIYers have and their circuit skills. For example on another thread I replied on earlier this week someone asserted they wouldn't have ground loops if power supply connections were made with twisted pair. Sure, if ya twist the wires hard enough they'll break and open circuit the ground, removing the loop, but at that point the supply doesn't work so well.

So I see a lot of elaborate isolation solutions without any measurements indicating the solutions are actually necessary---ESS gets 135 dBA DNR out of a non-isolated eval board, for example---with most of the problems seeming to be the result of use of unbalanced interconnects and power amps with excessive gain. A cheap balanced implementation with 1% resistors and a power amp set for unity gain (enough for about 85% of folks given a preamp on +-15V) has about 65dB more immunity to ground effects than default home audio implementations. It's not difficult to push that higher if there's need but since ground voltage swings in DC coupled consumer electronics are typically a few millivolts or hundreds of microvolts I've yet to hit a case where measurements have shown more rejection's needed. In comparison, isolating grounds tends to introduce bounce around half the mains supply if the free space capacitances are left to float---which is why plugging or unplugging throws sparks and why my experience is isolation tends to introduce more problems than it solves---and most folks won't reason through the circuit topology to understand which ground lift, interwinding, free space, etc. capacitances are charging from where and what the resulting bounces will be like.

Industrial applications can and do develop significantly larger ground offsets and I agree different solutions are called for in such cases. But the scenario for this thread is a tower/desktop/laptop with USB to a bus powered DAC which might or might not have a line out (the OP doesn't say but USB DACs are often used with headphones, so it's possible the only DC ground connection would be back up the USB cable). If we assume there is a line out to a pre or power amp that hardware's likely on the same circuit in the house wiring or maybe an adjacent one (and it's all single phase with relatively low currents). The ground topology in this configuration is the same as with a non-computer, DAC containing source like a CD player so the ground loops are also the same. A computer is a different load than a CD player so the currents in the grounds are different. However, they're not necessarily much different from having a non-audio computer plugged into the same mains circuit---if the audio grounds lie in the return path of the computer's draw they'll be in parallel with the wall wiring, will carry some of the current if they can, and will therefore bounce with the computer's load.

Isolating the computer-DAC link attempts to insert an all stop filter on that segment of the ground loops. Nominally, this requires the highpass corner from the capacitance across the isolation gap lie above the lowpass is the cabling RLCG with the amount of rejection provided increasing as the high and lowpass corners move farther apart. I've never seen an isolator intended for audio be characterized so I don't know how deep the resulting stop band is. Generally with these sorts of things it's not too hard to get 40dB but 60+dB usually takes pretty good filter design. But a 60+dB ground rejection improvement is pretty trivial to hit with balanced interconnects and some amp gain tuning.

Switching from an unbalanced to a balanced receive is a matter of adding two resistors to a layout. This costs about 25 cents in DIY quantities. Amp gain selection usually involves choosing different values of resistor and capacitors that would be in the feedback network anyway and is a zero cost change. In comparison, building an isolator for 25 cents would be a pretty neat trick and it still wouldn't offer the improved noise rejection of balanced or the SNR/DNR improvements of an appropriately chosen gain structure.

Don't get me wrong; I'm not saying not to use isolation. Just that it's probably more desirable to implement other ground management solutions first and then add isolation if performance requirements still aren't getting met.
 

sandyK

Member
2007-04-27 12:54 am
Sydney
metalsculptor
I don't know about other USB devices , but USB memory sticks have the 0 volts (black wire) internally connected to the screen wire.This is a potential earth loop .
Improved results can be obtained by fitting a 47 ohms .5W resistor in series with this black wire when used with an external +5V supply.. This has given improved results with USB memory sticks, and a member of another forum also reported improved SQ with his DACiT. In the case of USB memory sticks, even better results when used for .wav file storage and playback can be had with a low noise external +5V PSU, and the red and black wires disconnected at the PC end of the USB-A plug.The removed wires are replaced by a resistor , e.g. 220 ohms 1/2W across the terminals of the plug.Just be careful that it is insulated from the metal case of the plug.
DIY Audio member Erin from Melbourne suggested this modification and it works very well.
Regards
Alex
 
Isolating the computer-DAC link attempts to insert an all stop filter on that segment of the ground loops. Nominally, this requires the highpass corner from the capacitance across the isolation gap lie above the lowpass is the cabling RLCG with the amount of rejection provided increasing as the high and lowpass corners move farther apart. I've never seen an isolator intended for audio be characterized so I don't know how deep the resulting stop band is. Generally with these sorts of things it's not too hard to get 40dB but 60+dB usually takes pretty good filter design. But a 60+dB ground rejection improvement is pretty trivial to hit with balanced interconnects and some amp gain tuning.

Switching from an unbalanced to a balanced receive is a matter of adding two resistors to a layout. This costs about 25 cents in DIY quantities. Amp gain selection usually involves choosing different values of resistor and capacitors that would be in the feedback network anyway and is a zero cost change. In comparison, building an isolator for 25 cents would be a pretty neat trick and it still wouldn't offer the improved noise rejection of balanced or the SNR/DNR improvements of an appropriately chosen gain structure.
I agree that ground noise is manageable, I spend quite a bit of time doing just that, usually nasty sub uSec stuff with amplitudes well over a volt and earth paths an appreciable fraction of a wavelength at the noise frequency. I also agree that analogue isolation systems are expensive, complex and might not meet the performance required, OTOH the signal here is digital which is simple to isolate and imposes no performance deficits on the analogue side.
A few data points, one popular analogue isolation amplifier has 6 pf of capacitance across the isolation barrier, poor PCB layout could increase this considerably. This amplifier has a specified CMRR of 120dB, I forgot at what frequency.
Digital isolation does not have these issues, think optic fibre. To use an extreme example I could take a signal from a pulsed radar modulator deck (40Kv 0.5us rise time) and pass it out of its RF enclosure via a 4mm hole 160dB attenuation there then pass the fibre into another RF enclosure some distance away with a similar 160 dB attenuation.
Using differential (balanced)signalling helps, it sure would fix many of the hum and noise problems in a domestic environment and the silly thing is many power amplifiers have differential front ends which are converted to single ended. The chief disadvantage of differential signalling is that most differential amplifiers CMRR falls off rapidly with frequency.

@SandyK, that was the loop I was suggesting to break, putting 47 Ohm in that line would probably improve things provided the noise is not of sufficient amplitude to mess up the digital signal.
 

twest820

Member
2009-06-24 10:49 pm
Digital isolation does not have these issues, think optic fibre.
Sure, but this thread is specific to USB DACs (not SPDIF over Toslink) and USB 1.1, 2.0, and 3.0 are all copper interconnect standards. Optoisolated USB solutions, typically intended for medical applications, have been available for 15 years but audio oriented USB solutions are generally DC coupled. (As a historical aside, I was evaluating moderate cost, DC coupled USB audio interfaces in 1998 that were comfortably above 100dB SNR.) So optoisolation's typically done on the I2S link between the USB chipset and the DAC. If you have a look over in DIYA's digital line forum you'll find implementations of this, some fairly elaborate.

I don’t do much that’s discrete but a decent audio op amp exceeds 60dB CMRR below a several hundred kHz (see, for example, the typical performance graphs in the LME49710 and OPA1652 datasheets). Depending on layout and how much one chooses to spend on matching resistors this likely means CMRR performance of a balanced receiver is feedback network limited up to somewhere around a MHz. In an audio application that’s plenty of dead band to hand off to a lowpass before the op amp becomes the limitation. Most good designs will lowpass, either for RF rejection or to maintain stability in a composite power amplifier where the op amp’s providing a control loop for a chipamp or some power transistors. I would imagine this level of CMRR is more difficult to achieve in the input diff pair of a discrete amplifier as matching becomes rather involved. There’s limited need to do so in DIY, though; a fully differential op amp (such as the LME49724 or OPA1632) can swing for 50W RMS meaning amp drivers like the LME498xx family aren’t usually needed.

I’d not previously surveyed CMRR so hadn’t quite realized the advantage an op amp in the front end offers. I’d assumed 1% resistors (40dB CMRR) in my previous post so the numbers still hold, but not with as much margin as I’d thought. An LM3886 drops below 40dB CMRR at 50kHz typ, for example, and an LME49811’s 60dB CMRR point is probably around 20kHz
 
Thanks for the info on the I2S

My first post suggested converting the USB data to optical, regardless of the fact that USB specifies copper for the physical layer physical layers can be changed. Just like we convert ethernet to optical to get data off HV decks the rest of the network stack has no idea what physical layer is being used

If you want really good CMRR building an instrumentation amplifier out of high performance OP amps, is another option the problems with a single op amps CMRR is the impedance is not equal on both inputs, it might be possible to swamp the imbalance with resistors low impedance inputs are less noise prone anyway.
A low pass filter on the input is not a bad idea with noise being proportional to sqrt Hz it pays to only amplify the bandwidth the application requires.
 
So I see a lot of elaborate isolation solutions without any measurements indicating the solutions are actually necessary

I've yet to find a measurement which reflects the SQ problems caused by it well enough. But I do hear it so its definitely an issue - ears are the ultimate arbiter. I did have an intriguing issue in relation to isolation recently because I was using a class 1 amp (with mains earth) and lifting the 0V to chassis link improved the soundstage significantly. I didn't bother trying to measure this difference. I played around though and managed to preserve the soundstage with a custom wound inductor which maintained the 0V to mains earth link only at LF. But then on connecting my (class 2) DAC to this, all was fine so long as the DAC wasn't fed (over USB-S/PDIF adapter) from a mains fed laptop. Laptop on battery was fine, but with the mains adapter plugged in the noise was relatively massive (I documented this on another thread - obviously measurable). Now I'm using an amp powered from a laptop SMPSU (class 2) there's no issue at all.
 
Last edited:

twest820

Member
2009-06-24 10:49 pm
Yup, no measurements. I rest my case, your honor. ;) Somwhat more seriously, to the extent I've modeled such networks trafo leakage inductance often seems to be the dominant source of ground bounce. Leakage is rarely well characterized, though.

My first post suggested converting the USB data to optical, regardless of the fact that USB specifies copper for the physical layer physical layers can be changed.
Sure, but it's kind of a hassle as USB mixes differential and single ended signalling. Since USB offers power it's usually trivial to supply the USB chipset from the bus, moving one more noise to the digital side of the isolator and making things that much quieter for the DAC.

If you want really good CMRR building an instrumentation amplifier out of high performance OP amps is another option
Not as simple as it might seem at first glance; CMRR in an instrumentation amplifier is proportional to gain. In the unity gain case the amplifier's gain resistor becomes an open circuit, the input pair of op amps become buffers, and the CMRR ends up being controlled by the resistor matching of the third op amp's feedback network. This is the same as a single op amp with a differential feedback network, so the net result is more circuit for less performance. To the extent home audio systems need gain (and a good chunk of them don't) the optimum location for it tends to be in the DAC output buffers as this maximizes SNR and minimizes nonlinear distortion in the rest of the analog signal path. Nominally this means the rest of the signal path operates at unity gain.

That said, if one's building a power amplifier in a system with a gain structure such that a gain above unity is required for the power amp using a instrumentation amp type front end allows a bit more performance to be gotten out of the circuit if one doesn't wish to go so far as to build a composite amplifier. This happens, for example, when the power amp needs to swing beyond +-28V or so (which is pretty darn loud unless you're trying to use a resistor as a speaker). Hypex does this in their UcD and nCore amps, for example, presumably because maintaining stability of an op amp feedback network wrapped around a class D modulator tends to be a bit tricky.
 
Not as simple as it might seem at first glance; CMRR in an instrumentation amplifier is proportional to gain. In the unity gain case the amplifier's gain resistor becomes an open circuit, the input pair of op amps become buffers, and the CMRR ends up being controlled by the resistor matching of the third op amp's feedback network.

Yeah I try to avoid using that architecture in a unity gain configuration, its way better suited to gains of 100 or up. CMRR also degrades if there's any source impedance mis-match too. Have you come across AD830 though? An entirely different beast, not based on the traditional opamp architecture, but something infinitely more interesting. I recently found NXP has some parts based on the same Barrie Gilbert architecture. The AD830 still manages 100dB or so of CMRR at unity gain (and impressively beyond audio freqs too) but the source impedance still matters and is likely dominant in the real-world performance.
 
Yeah linearity is an issue on paper, yet in digital audio (subjectively) it works better than parts that on paper have better linearity. Just goes to show how measurements can be deceptive :D

I suspect the parts I referred to are obsolete, datasheets are available and they do show up on Taobao. TDA8575, TDA8578/9 come to mind. Their chipamps for car radio also appear to be based on the same architecture - they do sound extremely good.
 
But I do hear it so its definitely an issue - ears are the ultimate arbiter.

Yes, well, it'll definitely be an issue if you're hoping to sell one to abraxalito.

The real issue could be, why are you designing and building this DAC?

If you intend to sell them, or hope that people will buy boards and build them, then you'll have to pander to all the audiophile phantasmagoria under the sun, or at the very least provide the objectivist buyers well-founded specs, probably supported by test results generated using professional quality test equipment. Speaking of which, have you seen the nwavguy's ODAC?

NwAvGuy

If you just want to listen to it yourself, you'll probably be satisfied to design it to meet a reasonable specification, but if you want other people to want it, then you might be better to design it to meet as unreasonably demanding a standard as you can possibly imagine, since there appears to be no level of technical achievement or measured performance guaranteed to meet with the approval of the golden eared.
 
The real issue could be, why are you designing and building this DAC?

Wow, as they say 'speak of the devil'. I just mentioned you over on another thread. Simple answer to this question - 'for fun'.

If you intend to sell them, or hope that people will buy boards and build them, then you'll have to pander to all the audiophile phantasmagoria under the sun, or at the very least provide the objectivist buyers well-founded specs, probably supported by test results generated using professional quality test equipment.

I've no aversion to measurements but no I won't be providing 'proper' measurements where NwAvGuy gets to define what 'proper' means. Measurements made with my portable Sony recorder though and shown on Audacity suit me. If anyone needs more 'professional quality' then I can do without them as 'customer'. :) I don't intend to sell them myself, I prefer the ARM business model.

Speaking of which, have you seen the nwavguy's ODAC?

Yes, studied that and his O2 amp too. Which incidentally was the inspiration for the name of my DAC which is 'Ozone' (O3, geddit?). He makes unsubstantiated claims for its transparency though, and I took him to task for this before his 'disappearance'.

If you just want to listen to it yourself, you'll probably be satisfied to design it to meet a reasonable specification, but if you want other people to want it, then you might be better to design it to meet as unreasonably demanding a standard as you can possibly imagine, since there appears to be no level of technical achievement or measured performance guaranteed to meet with the approval of the golden eared.

The golden eared buy on SQ though. They're my prime target customers.
 

twest820

Member
2009-06-24 10:49 pm
TDA8575, TDA8578/9 come to mind.
TDA8579 is still in production. It'd be nice if the datasheet showed more of the feedback implementation but, eh, back burner project to figure out the innards I guess.

Difference amplifiers like the AD8270/8271, LT1995, and INA134/137/154/2134/2137 are worth a look. Quite a bit more linear than the TDA8579, generally higher CMRR, reasonable noise levels, and thermal tracking that's quite a bit more expensive to achieve with discrete resistors (put a few volts RMS across a 1-2k 0805 or 1206 0.01% 5ppm and the accuracy is likely to halve). For cases where one wants CMRR beyond what 0.5 to 0.1% resistors can deliver I would probably default to the INA154 despite the INA family's low +PSRR requiring a good reagulator.
 
Had a quick squint at AD8270 - don't buy the 145dB distortion claim, also I reckon it won't have linearity where it matters for audio - which is low level IMD in the presence of OOB signals. LT1995 though looks more interesting and is reasonably priced but I reckon LTC6552 holds more promise for audio and is slightly cheaper. I'm going to avoid INA154 as the only clue to it being linear is measured at 10VRMS output - hardly audio territory ;)
 
Last edited: