Measuring DAC Accuracy

Holy Schitt! You guys are nutts.

 

Yeah. The biggest glitch is at zero transition and there will be smalier ones at specific transitions like 16384 to 16385. Note that the S19 has glitches too... with less of spike, but seemingly at every transition.

 

Looks great @atomicbob, seems like less DC block filter issues there.

LOL! It takes days for the MB DACs to settle?! Kool! That correlates highly with what folks are saying about leaving their gear on for best results.

 

@atomicbob: can you show DS DACS results with 16 bits?

 

BTW, just wanted to show you guys different levels of issues using a 200 Hz pyramid like signal as it goes through an high pass (DC block) IIR (5th order Butterworth) with different corner frequencies (note a servo is still an IIR filter, with maybe a very low corner frequency):

1000 kHz (this is not only a non-linear phase cluster f**k, the signal is in the stop band of the filter, i.e. it's gone)



200 Hz (right at the signal)



20 Hz



2 Hz



0.2 Hz



No high pass IIR



So, the corner should be in the fractions of Hz for the phase distortion to stop screwing up all the way to 200 Hz.

Using 16 bits may or may not show the bowing deal. But it would likely be an indication of how low the DC block is in frequency, or if there is a DC block at all.

This IMO is a trade off. Somewhere in the chain there must be a DC block to avoid DC damage to the drivers. If its not in the DAC, it's going to be in the amplifier. And depending on the corner of the high pass, one may forget about non-bowing and the kind of accuracy we are discussing here with these tests. Cuz the amplifier just undid it.

Ideally one wants a very low corner frequency in the DC block high pass. In the analog domain this requires large resistors and/or large capacitors, or maybe a servo which exploits the high impedance of the non-inverting input of an op-amp (or similar). Large resistors are bad because they inject noise. Large caps may be expensive if one wants low distortion, and huge either way. Using op-amps may necessitate the need of one that has low offset current while at the same time being low distortion and low noise. Many such parts with the super-awesome specs are not cheap and may only be available in SMD. There are OK through-hole parts. But they are in the OK range compared to the SMD parts.

In all cases, one still has an IIR. Best one can do is move the corner as low as possible. Which may also mean that it could take time for things to settle on turn on and off (transients).

(BTW, I'm very much open for corrections if I said something wrong or may be misguided in my thinking above)

 

I guess Schiit should have dithered the 20bit output to make JA happy.

--

I took at the peek of the output wave files from ARTA and didn't like what I saw, so I decided to write a program to produce exactly the output I desired.

• 96kHz
• 750Hz triangle
• Values from -32 to 32 in up and down EXACT pattern: 0xFFE0, FFE1, FFE2 ... FFFF, 0000, 0001 ... 001E, 001F, 0020 ...

Instead of going for 32+1 possible values like before, we will go for 64 + 1 (zero). Basically looking at the performance of the 6 LSB bits of the DAC.

Then I wrote a program to analyze the output. I want to see how the sampled levels were being distributed at each point. Below are some preliminary results for 25 cycles. Whiter dots indicate areas of higher concentration.

A-GD S19


Laptop Output


Modi Multibit


Modi 2


Results are interesting. Is there such a thing as amplitude jitter?

I'll do some more analysis of the data later.

 

I'll wager it is noise if it's all random looking. If there is a pattern I would call it distortion.

 

Well, we already know the S19 has a lot of distortion. Is this distortion a result of the analog circuit, digital converters, or both? Or maybe lower frequency stuff underneath?

 

I dunno much about the S19, but here is a convenient linky to their specs:

http://www.audio-gd.com/Master/Singularity19/Singularity 19EN_Specs.htm

They claim it's a "Dedicated Discrete R2R DAC", which to me means they did something custom. Looking at their DA-M1 cards (assuming those are the actual DACs), I see what looks like a small & cost effective FPGA and a shit load of resistors, and then some more shit loads of resistors in another different module. Resistors may generate a bit of noise, so I would be concern about those.

When looking at stuff in the time domain, and perhaps to check jitter and noise roughly, one can look at an eye pattern. Some folks are used to seeing "1's and 0's" (BPSK & QPSK) in those, but in reality, they are used to check things out on amplitude and quadrature amplitude modulation which have several levels. Depending on how "closed" the eye pattern is, one can get a feel for how shitty things are, both on timing and noise. It may not give you a lot more than just checking the spectrum of the output signal vs. the input signal for audio applications if trying to just differentiate between noise and distortion though.

However, the eye pattern might give you a relatively good feel for noise/distortion in amplitude (y-axis noise) vs. jitter (x-axis noise).