Aug 5, 2016 at 8:42 AM
- Feb 11, 2008
- Reaction score
- Feb 11, 2008
For those of you who don't frequent the DAVE DAC thread, this was posted earlier, by Rob, and makes interesting reading, even for Mojo owners:
This thread got a little interesting last week with talk about the PSU within Dave. Now that the dust has settled, I thought it would be a good idea to discuss the thinking behind the power supply with Dave.
When it was mentioned that Dave's PSU was just a simple medical SMPS I had a little chuckle to myself.
Because the medical SMPS is not Dave's PSU - at least not the PSU that actually matters. You may as well talk about which power source you are using from the mains - nuclear, fossil or wind/solar? OK that's an extreme exaggeration, but power is simply a serial chain, and the medical SMPS is not the critical component within Dave's power structure. It's actually an extremely elaborate and complex interdependent system. And this is the reason why I found the post amusing. So that you can understand, lets talk about what you need from a PSU within a DAC:
1. RF noise. This by far is the most important thing in a PSU and analog electronics; it's something I talk about a great deal. The problem is that analogue audio components are non-linear, and very non-linear at RF (and by that I actually mean 20 kHz to several GHz). When a random RF noise gets into active audio components, it distorts with the wanted audio signal and creates inter-modulation distortion - and some of the inter-modulation distortion has an audio component - audible random noise. What happens with this is when there is no audio signal, you get no inter-modulation distortion, hence no extra noise. When the audio signal increases, the inter-modulation products increases, and noise goes up - so noise levels becomes linked to the signal level and you get noise floor modulation. Now this issue is easy to measure, and taking steps to remove RF noise lowers noise floor modulation. Additionally, you can improve the analogue electronics open loop RF linearity too.
The problem with noise floor modulation is the ear/brain is extremely sensitive to it, and certainly can detect levels of noise floor modulation that is below the ability to measure. My own tentative conclusions (or rule of thumb) are that one can hear levels of noise floor modulation down to -200dB - currently we can measure noise floor modulation at -180 dB, and Dave has zero measured noise floor modulation. In terms of SQ, if noise floor modulation is say around -120 dB (typical class D) you get considerable hardness and glare; at -140 dB its grain in the treble; below -160 dB then things sound much smoother with better instrument separation and focus. This continues until about -200dB (and perhaps even lower - reducing RF noise is not something that has an acceptable limit).
But to get this level of performance required me to do many things; and part of that was the PSU system. I call it a system because it is lots of parts working together, with sources of noise within the DAC contaminating other parts. One source is the mains power, so this is dealt with a filter that starts at a few 100 Hz to several GHz - each PSU line (+15v,-15v,+5v) from the SMPS is individually filtered with a complex multistage filter. That covers RF noise initially, but every analogue part is individually regulated and RF filtered again. Moreover, each digital module is individually RF filtered too, as each digital part of a DAC is a huge RF noise generator.
2. PSU impedance. The impedance of the power supply is crucial too as it can create distortion which is audible. The actual mechanism for distortion from an amplifier is due to the fact that an amp draws signal related current from the PSU; this current then creates a voltage drop in the power rails (an error) that is signal dependent; this error then is fed back to the amplifiers output by the amp power supply rejection ration (PSRR); the error then creates distortion in the output. For Class A it is second harmonic, for class B it is very serious HF harmonics extending to infinite harmonics. Now its very easy to design the amp to avoid this problem; simply use low enough output impedance PSU and an amp that has a large enough PSRR. It is easy to calculate your requirements; much more difficult to design a stage that meets those requirements. So with Dave I wanted no measurable effect from this; this meant a PSU impedance in the OP stage of 3 milli ohms, which would make this effect un-measurable at -180dB. And that's exactly what I achieve; loading the output stage shows no measurable PSU induced distortion at all.
3. Reference supply. So far I have talked about the requirements for the analogue section; and this is an interaction between amp topology (sensitivity to RF and PSRR) against the actual levels within the PSU. So you could use a poor PSU with an amp that had amazing RF and PSRR or vice versa. My approach with Dave is to use both strategies - get the best innate isolation with the best filtering I could do. But there is a PSU where whatever happens on the PSU will be directly on the output and this is the reference PSU. Now nobody talks about the reference for two reasons - it is buried inside a DAC chip, and because people are not aware of how crucial this component is. The reference supplies the voltage that is used to create the analogue voltage (or current) output on the DAC, so clearly if that voltage changes it will have an immediate and 100% impact on the output. Normally, with a silicon DAC the reference presents a problem; noise on the reference will appear on the output, so making the SNR and dynamic range (DR) depend on the reference. Its actually extremely difficult to have a 120dB reference voltage on chip; so to overcome this they use a differential structure (balanced + and -). This means the reference noise, when its reproducing small signals is cancelled as it becomes common mode. So you get good DR figures from a poor reference. But when the signal becomes larger, the cancellation stops and you then see the reference noise and you get noise floor modulation - which is very measurable and very audible. Its one reason why chip based DAC's have large amounts of noise floor modulation.
Now with all my DAC's I have the freedom to design my own discrete reference, so can use very low noise references - and I do not have to worry about substrate noise upsetting the reference, something that is impossible to eliminate on chip DAC's. So the reference is crucial for noise floor modulation, but its also crucial for distortion too. Any signal related currents on the reference will create an error voltage, which will modulate the output causing distortion. In the case of a differential structure it causes amplitude modulation - a problem that won't be seen with simple measurements of distortion. With my SE designs I get to see it directly, which means its easy to measure it and solve it - and because you are also eliminating amplitude modulation (which is very audible) it gives me a short cut to better sound quality.
With Dave I had to go to immense efforts to solve these problems - and its one reason why I get such good measured performance. But it took many years to design and get right. With Dave the reference OP is less than 0.1 milli ohms at the feedback point (that's a damping factor of about 100,000) and at the PCB power plane each flip flop sees only 0.5 milli ohms. Now the actual distortion you get with reference impedance does depend upon the flip flop currents, and this is a system design issue of balancing resistor noise with distortion - resistor noise being the dominant noise source for Dave. To eliminate measurable noise floor modulation I had to design the noise of the reference to be less than 0.3 uV 20 to 20k. Considering it is a 5v voltage, that is a DR of 144 dB. What other power supply or amplifier has 0.1 milli ohms and -144 dB noise?
So you can see my amusement about the PSU discussion. Would using a better SMPS give better SQ? Maybe, but the evidence suggests no. When I used this power supply, I initially used the DAC 64 unit. But I could see switching noise at -130 dB at 40 kHz - going to the better medical PSU eliminated this problem - but when I replaced the unit I could not hear a change (I did not do an AB test, just plugged it in and listened henceforth and I was not struck by my system sounding better). So far the listening evidence is that is good enough.
I always get amused by audiophiles listening with their eyes, so a big PSU with lots of decoupling caps must be better than a small PSU. But the actual reality is that PSU interaction and sound quality is a hugely complex interdependent system problem, and eliminating these problems are not visible, nor are they easy. Also, the techniques I use that I have discussed in this post is used from Mojo up to Dave; its just a question of how much budget I have available to solve these issues. Even Mojo has no measurable noise floor modulation at -170 dB; and it uses the same principles of a discrete reference circuit - albeit not as sophisticated and costly as Dave's.