Return-to-zero shift register FIRDAC

[MarcelvdG]

The bottom solder mask was missing from DAC3filter.zip, it's corrected in the attachment.

 

[nautibuoy]

Watching with interest. I had a look at the gerbers and it looks doable now I've started to get confident with smd soldering.

 

[MarcelvdG]

I've ordered the boards and the components, but the components won't ship until February. I've changed the OPA1678's into NE5532's and used whatever brand was available for the 74xx chips, otherwise it would at least take until September.

To be continued in March or April.

 

[nautibuoy]

I've ordered the boards and the components, but the components won't ship until February. I've changed the OPA1678's into NE5532's and used whatever brand was available for the 74xx chips, otherwise it would at least take until September.

To be continued in March or April.

It's good to see this progressing Marcel, those few months will soon pass and I'm sure there is a small group of us awaiting your prototype build with great anticipation.

Seasons greetings and happy New Year to you too...

 

[MarcelvdG]

Same to you and to everyone! I hope next year will be a better year than this year, but I hoped the same at the end of 2020.

 

[nautibuoy]

Hi Marcel, a little something to occupy you while you're waiting for parts if I may. 🙄

I've been catching up with Raj1, who was part of the group buy for the DSD Valve DAC - he's progressing that project and you may recall he was planning to use the 'CD Enhancer' on the output of the Valve DAC, in place of your reconstruction filter section.

http://itsonlyaudio.com/ultranalog/cdenhancer/cdenhancer2_3a5.pdf
Raja is kindly sending me some transformers that suit the CD Enhancer and I have the rest of the parts to hand to build it - just wondering about using it with the RTX FIRDAC in lieu of the op-amp output filter stage? There's no urgency, obviously, but it would be good to hear your thoughts on the possibility.

 

[MarcelvdG]

If the circuit has been designed for a zero DC common-mode input voltage, then you will have to increase R7 to 220 ohm or 226 ohm (if you prefer E96 values) to adapt it to the 1.25 V common-mode voltage that will come out of my DAC PCB. With that change, it should work. The noise floor and distortion of the DAC could be somewhat worse than with a virtual ground load because the current drawn from the voltage reference will be more data-dependent, but I have no idea how big that effect is.

 

[nautibuoy]

Thanks Marcel. Probably worth a try then but all in good time.

 

[MarcelvdG]

Update: as things were going out of stock faster than getting into stock, I asked Mouser to send the stuff they had. As a result, I could already populate most of the DAC board and do a simple functional check of the voltage reference. Using the BCM56DS transistors because the NST45011 won't be in stock for a very long time and using 1 kohm for R₄, I get 4.96 V at the output of U_1A. So the bandgap starts and gives the right voltage, now I have to figure out a way to measure its noise.

 

[gentlevoice]

Hi Marcel ... Have been following this with interest on the side - I do look forward to seeing how it progresses.

Cheers - Jesper

 

[Raj1]

Yeah, this is exciting. I'll probably be down for one of these also.

 

[nautibuoy]

I feel another group buy down the line, perhaps this time, the supply situation permitting, for factory assembled boards.

 

[Markw4]

Marcel,
Just wondering if you have seen the discrete dac implementation in Marantz SA-10 player? Schematics are in the publicly available service manual. Looks like a CPLD drives two multi-channel D-Flip Flops per dac channel. Also kind of looks like ground bounce and similar transients are somewhat canceled out by use of the two FF chips and driving them differentially. In addition, the FF outputs have large-ish series resistors so that output array could be used a maybe a multi-bit oversampled dac or else a 1-bit dac depending on how the CPLD is configured. Interesting design...

 

[MarcelvdG]

No, I haven't. It's indeed interesting. It could also be differential to make the load on the reference less data-dependent, especially if they use RTZ (which again depends on the CPLD).

 

[MarcelvdG]

Measured reference noise at the output of U_1A, so at the reference amplified to 4.96 V:

Wideband (up to 44 kHz?): 29.784 nV/sqrt(Hz)
Around 25 Hz: 47.866 nV/sqrt(Hz)
Estimated 1/f corner point: 39.57 Hz

I measured the noise by connecting the output of U_1A to pin 2 of the microphone input of a Fostex FR2-LE field memory recorder via a 47 uF bipolar electrolytic capacitor and a 47 ohm resistor, connecting ground to pin 3 via 47 uF and 47 ohm, also connecting ground to pin 1 of the Fostex and making 30 second recordings of the noise of the reference, the noise with R₂₃ removed and the noise with the RC network to pin 2 of the microphone input grounded (to be able to subtract the noise of the FR2-LE itself). The Fostex was set to 96 kHz sample rate, 24 bit, phantom supply off and was running on batteries. Nearby cellphones were off.

With R₂₃ removed, the noise is easily calculated and fairly accurately known, as it is largely dominated by the thermal noise of R₂₀. That is, I use R₂₀ as a kind of calibrated noise source for the reference noise measurement.

I measured the RMS levels with the GoldWave audio editing program, using the volume-maximize function first and then volume-match. I've done that over the full frequency range of an FR2-LE at 96 kHz sample rate, presumably about 44 kHz, and narrowband in a band from 20 Hz to 30 Hz, using the GoldWave bandpass filter function (steepness = 20). I didn't use the first five seconds of the filtered signal to give the filter plenty of time to settle.

 

[MarcelvdG]

For comparison, ratio of the white noise and the DC voltage for various voltage and current references:

85A2 (glow discharge): 648.8 nV/sqrt(Hz) and 85 V, so 7.633E-9 Vref/sqrt(Hz)
LM723 (Zener / avalanche diode): 86 uV from 10 Hz to 10 kHz at 5 V output voltage, so 172.1E-9 Vref/sqrt(Hz)
LM317 (bandgap): 0.003 % of Vout from 10 Hz to 10 kHz, so 300.2E-9 Vref/sqrt(Hz)
uA7805 (bandgap): 40 uV from 10 Hz to 100 kHz at 5 V, so 25.3E-9 Vref/sqrt(Hz)
LT1236 (buried Zener reference): 2.2 uV from 10 Hz to 1 kHz at 5 V, so 13.98E-9 Vref/sqrt(Hz)
LT3081 (unknown, presumably bandgap): 5.7 nA from 10 Hz to 100 kHz at 50 uA, so 360.5E-9 Iref/sqrt(Hz)
LT3042 (unknown, presumably bandgap): 6 nA from 10 Hz to 100 kHz at 100 uA, so 189.7E-9 Iref/sqrt(Hz)
TL431 (bandgap): 125 nV/sqrt(Hz) at 2.495 V, so 50.1E-9 Vref/sqrt(Hz)

This reference (bandgap): 29.784 nV/sqrt(Hz) at 4.96 V, so 6.005E-9 Vref/sqrt(Hz)

Pretty good, whether that matters at all is a different question.

 

[EUVL]

TL431 Onsemi ist 48nV/sqrt(Hz).


Cheers,
Patrick

 

[Mark Johnson]

Marcel, why not add some brute force post filtering on the reference? Either active or passive or (e.g. 3rd order LP) both?

 

[MarcelvdG]

I use a first-order low-pass, 4.99 kohm and 47 uF, in fact that's why I wrote about the reference noise: "whether that matters at all is a different question." I just wanted to try this type of bandgap circuit ever since I read about it in an old issue of Electronics Letters.

 

[MarcelvdG]

The measured noise is without filtering, of course.