Improving a cheap chinese PCM1794 DAC

It's not...it's calculated.Measuring an ess9018 dem board with an AP system 2 there was no measurable distortion at 1khz...It was like there was nothing there...

This is not correct.

Use a pre ADC input notch filter - those distortions are very well measured
by notching out the fundamental and looking at the remainder.

Even without using a the notch filter, AP can get down to harmonics of
around -130dB. Check out Stereophiles measurements of DAC's like Chord
Dave and various Dcs results.

WRT ADA4898 current noise:

I suggest you re-do noise calcs of ADA4898 and you will see that it -does-
have lower noise, even despite high current noise. Do I need to school you
on current noise calcs? :)

WRT fast edges being transferred directly through I-V opamps FB resistor:

I suggest you fire up LTspice and feed fast current edges into audio speed
opamps, even using a substantial FB cap, the fast edges are NOT transferred
directly to the OP but are ameliorated with ringing where the opamp can't
initially deal with the edge rates.

WRT vid versus audio opamps:

Video opamps have greater speed, audio opamps have lower distortion due
to higher OLG at audio freq. To achieve most modern DAC's low distortion
capabilities, you generally need to use an audio opamp.
The ADA4898 is a bit in between having combination of low distortion, high
OP drive and fairly fast slew rate.

T
 
I think that i listened to the transducer...some esl headphones plugged into an audio amplifier with 100x higher distortions than the DAC itself...So i assume that it was impossible for me to listen to the DAC distortions or noise...I'm more into masking techniques. I prefer to replace nonlinear distortion and patterned noise with harmonic distortion and unpatterned noise...

Can we get this thread back to discussing PCM1794 DAC / board.

These are great DAC's for DIY'ers to play with being completely hardware
configurable, also having a high voltage compliance, hiZ current OP makes
them great for all sorts of DIY I-V and analog stages.

T
 
Account Closed
Joined 2010
I often find that it's not the dac who really is to be blamed for a bad sound.There's no rule for this game .A good recording engineer can make a good sound out of very poor equipment and learning how it's done won't harm anybody.
On the other hand buying overly expensive dacs that are able to read overly processed and upsampled audio formats made out of the original 16 bit WAV which again was made out of a poor master tape....that's hilarious ! It happens to me that i mostly like the music of 70's...90's where most of the recordings were analogue, so any later digital recording is actually a manipulation of some sort of the original noisy and heavily distorted material...
 
Account Closed
Joined 2010
This is not correct.

Use a pre ADC input notch filter - those distortions are very well measured
by notching out the fundamental and looking at the remainder.

Even without using a the notch filter, AP can get down to harmonics of
around -130dB. Check out Stereophiles measurements of DAC's like Chord
Dave and various Dcs results.

T
2700 Series Audio Analyzers - Audio Precision(C) | The Global Leader
http://www.testequipmenthq.com/datasheets/AUDIO%20PRECISION-SYS-2722-Datasheet.pdf

i see a minimum residual of -112 db THD+N at 1khz so you can't get lower than this

If you know better what i saw with my own eyes then...I'm all ears, GOD!
 
Account Closed
Joined 2010
This is not correct.



WRT ADA4898 current noise:

I suggest you re-do noise calcs of ADA4898 and you will see that it -does-
have lower noise, even despite high current noise. Do I need to school you
on current noise calcs? :)

T
Please do it for me, i have no university background of any sort, i only thought that TIA is transforming the current noise into voltage noise in the most linear way.
 
...On the other hand buying overly expensive dacs that are able to read overly processed and upsampled audio formats made out of the original 16 bit WAV which again was made out of a poor master tape....that's hilarious !...

Enjoy a good laugh on me. No problem.

But, I think if you ever heard well recorded CD on something like Benchmark DAC-3 or Chord Dave (different from each other, of course) you might stop laughing and start saving up for one. You may recall that properly dithered 16-bit recordings made from master taps contain information below the 16-bit level, it is just mixed in with some noise. Fortunately, human brains are good at extracting signals from noise, so there are details to recordings you have no idea even exist. Regarding the cost of great dacs, I completely agree with you that they cost too much. My work in this area is to make it publicly known how how one might make a dac that could compete with some of the best. If more dac manufacturers knew how to do it, hopefully we would eventually start to see more competition and lower prices.
 
Last edited:
Account Closed
Joined 2010
This is not correct.


WRT fast edges being transferred directly through I-V opamps FB resistor:

I suggest you fire up LTspice and feed fast current edges into audio speed
opamps, even using a substantial FB cap, the fast edges are NOT transferred
directly to the OP but are ameliorated with ringing where the opamp can't
initially deal with the edge rates.


T
Hopefully 60 degrees phase margin of LM833 can handle it...What do you think? On the other hand i prefer to have a limited set of rules when assessing a complex problem even though i know i'm going to ignore real problems .I wasn't against using higher slew rate op-amps , just against ignoring lower output drive.
 
Account Closed
Joined 2010
Enjoy a good laugh on me. No problem.

But, I think if you ever heard well recorded CD on something like Benchmark DAC-3 or Chord Dave (different from each other, of course) you might stop laughing and start saving up for one. You may recall that properly dithered 16-bit recordings made from master taps contain information below the 16-bit level, it is just mixed in with some noise. Fortunately, human brains are good at extracting signals from noise, so there are details to recordings you have no idea even exist. Regarding the cost of great dacs, I completely agree with you that they cost too much. My work in this area is to make it publicly known how how one might make a dac that can compete with some of the best. If more dac manufacturers knew how to do it, hopefully we would eventually start to see more competition and lower prices.
I can't argue about everything.Actually i tried to tell a lot of people to be mindful of the mechanisms that lie behind noise perceptions.One of the most obvious one is the fact that if the noise has no pattern and the distortion is particularly harmonic we will hear "better and detailed" music out of fabricated signals by your system. When playing a master tape on the most expensive reel to reel machine , you can't get less than 0.5% distortion out of it but you'll like it as no other.No matter how low distortions your dac might have , the rest of your audio chain will make it irrelevant which makes me think that those dac manufacturers don't tell the whole truth about their toys and they do fabricate distortion and noise so that the rest of your equipment become irrelevant and buy their toys.
 
Last edited:
The best dacs will sound more like the master tape than any other way. It sounds very, very good. And of course, the rest of the system need not obscure some of the details coming out of the dac. Right now I use a Benchmark AHB2 power amp, the lowest distortion I could find and SOA, and the speakers I use are the same as was used to mix many hit records, augmented with sealed subs. It works well enough to hear everything coming out of the dac, although it could be further improved in some ways.

Again, and unfortunately, good audio remains expensive (although huge amounts can be spent for rather poor audio too). Good news is that Bruno Putzeys is starting a new company to produce lower distortion switching amps that ever before, and with ultra-low IMD. They will probably be expensive at first, but over time prices will probably come down. These are times we live in right now, we still don't know how to make good audio at a cost most consumers are willing to pay. May not stay that way forever, hopefully not.
 
Last edited:
2700 Series Audio Analyzers - Audio Precision(C) | The Global Leader
http://www.testequipmenthq.com/datasheets/AUDIO%20PRECISION-SYS-2722-Datasheet.pdf

i see a minimum residual of -112 db THD+N at 1khz so you can't get lower than this

If you know better what i saw with my own eyes then...I'm all ears, GOD!

Obviously you need help using google to find stereophile measurements.
Chord Dave:
https://www.stereophile.com/images/617Davefig08.jpg
Dcs Vivaldi:
https://www.stereophile.com/images/114dcsDvivFig16.jpg
 
Please do it for me, i have no university background of any sort, i only thought that TIA is transforming the current noise into voltage noise in the most linear way.

Well why go sprouting something when you don't understand it?
You don't need a degree, just a willingness to learn.

Current noise =2.4pA/rtHz x FB resistor value, say 850R or so.
2.4x10^-12 x 850 = 2nV/rt Hz.

Thats less than half the voltage noise of 5534 if used for 1794 (unity gain)
If used for Sabre the resultant voltage noise (from current) will be less.
 
dreamth,
Suggest you look around for a copy of, The Art of Electronics. It covers the kinds of things Terry is talking about and much more, and it does it without getting too theoretical, its more of a practical electronics design book. Probably the best of that type.
 
Hopefully 60 degrees phase margin of LM833 can handle it...What do you think? On the other hand i prefer to have a limited set of rules when assessing a complex problem even though i know i'm going to ignore real problems .I wasn't against using higher slew rate op-amps , just against ignoring lower output drive.

I don't think it's a case of phase margin.

This is the OP of PCM1794 as referred by by abraxalito in this thread
http://meandr.org/wp-content/uploads/2015/07/153.jpg
http://meandr.org/wp-content/uploads/2015/07/171.jpg
Article here:
Какой звуковой ЦАП лучше? — Меандр — занимательная электроника
As you can see in last pic, the edge rates are really fast, as would be
expected fom CMOS logic in the DAC, and there is also a lot of ringing. I
don't think trying to pass this 'through' an opamp via the FB resistor is a
good idea. LTspice shows low slew rate opamps to behave terribly and even
faster ones not that well. We haven't even considered real world layout and
various non ideal component inductances etc etc.

Fixing the problem at the source, ie; straight after the DAC makes a lot of
sense to me. But it doesn't integrate well with all DACs, the 1794 is an
exception due to it's healthy voltage OP compliance.

Please be a little bit more constructive here. The thread is tweaking a
PCM1794 board. These DAC's are fairly cheap and capable of really high
performance if treated right.



T
 
Hey, thanks for the interesting discussion. I have chosen the PCM1794 after I read this article, which has in depth measurement/analysis/comparison of delta-sigma/multibit DACs and their differences. You may find it interesting: Какой звуковой ЦАП лучше? — Меандр — занимательная электроника. It is in russian so you should probably use Google Translate.

By the way, here is the whole setup in a temporary enclosure out of a Polo RL scarf :zombie: until I find someone with a CNC nice enough to make me two simillar enclosures.

raspberry_volumio_xmos_pcm1794.jpg


The setup consists of:
- A DIY AC filter
- 2x15VAC+2x9VAC R-Core transformer feeding the DAC board
- 2x7,5VAC feeding the XMOS reclock and Raspberry Pi
- Raspberry Pi 3 running Volumio fed by a LT1083 + CRC filter PSU(on the black prototype PCB, painted black with a spray can)
- JLSounds XMOS, with it's reclock after galvanic barrier fed by an LM317 + CRC PSU(also on the black board)
- The modified chinese PCM1794 board
- An LCD display+MCU+rotary switch that came with the chinese board used to switch digital inputs and display current signal frequency

While still burning in, the sound of the PCM1794 is amazing. TBH I have never expected such a detailed sound that is still warm and pleasant to listen to from a delta-sigma DAC except for the most expensive Sabre pros. I don't believe the TDA1541 myths anymore.

The good thing is that the whole streamer/DAC costed less than 250$ including the trafos, raspberry, xmos, memory card, DAC board and all the components for the upgrade. It allows me to listen to music from online streaming services such as Spotify and Tidal, local HDD and I can still switch to the toslink coming from my TV with one click and use the DAC for Xbox(pass-through) and Netflix.

I still have to add the IR receiver for an Apple remote I have laying around but Volumio has a web interface so it's not mandatory.

Pachy

Getting back to original thread subject, it's worth going through the DDDAC
thread. There are a lot of tweaks to get the best out of 1794.

A NOS 192/24 DAC with the PCM1794 (and WaveIO USB input)

There's a lot to go through but a few things that were implemented:

a/ increasing OP current through current scaling resistor
b/ Using a CCS instead of that resistor for better noise immunity
c/ Increasing analog voltage supplies.

cheers

T
 
Account Closed
Joined 2010
dreamth,
Suggest you look around for a copy of, The Art of Electronics. It covers the kinds of things Terry is talking about and much more, and it does it without getting too theoretical, its more of a practical electronics design book. Probably the best of that type.


I don't have copy...i have the actual third edition which i bought in late 2015.
I spent a few weeks measuring prototypes based on the most famous chips of the time in 2015 on both system 1 and system 2 , including opa1612, opa1642, ne 5532, ad797, njm4562, lm4562, lme49880, opa2132, etc... and measurements aren't that optimistic as Terry would like them to be. They also vary with sample and batch as the Typical value is rarely met. I'd guess that the dev boards sent by the dac manufacturers have very well selected op-amps for the job.

The datasheet shows a TYPICAL 2.3nA current noise.The values for that resistor may vary from dac to dac and actually you might get worse results than you think with new components. If you add the voltage noise with the one that come from In X Rf you might simply get the same results as the old ne5534 or just a bit lower ...OPA2132 shows double noise than opa1642 while the reality is exactly the opposite, where the same TIA based on opa3132 has AUDIBLE lower noise than opa1642, the THD for a TI ne5532 is lower than opa2132, but it varies substantially with temperature and time, while opa2132 has a higher, but very stable THD and so on...


I'm still open for schooling though :)


I started to simulate my own I/V stage with a complex input but I'd like some help for the parallel current sources timings.I don't exactly know how to sim a real dac output so i attached some photos for my spice sim. If you think i got it all wrong please tell me what to do! Please make this I/V of mine sim really bad!



I just launched an order for ada4898 and many other components to build the best variant of my own i/v , so that i can test them along with my own toys. For now I only have a Phillips based on TDA1543 plus LM833 and transistor front end I/V and a Sony based on a tda1541 , 4x oversampling and no output filter(SONY CDP750 as stock!!!) to be modded with the best current converter money can buy, so theoretically we should have a 16 bit nightmare there, isn't it?
In the end I won't be convinced that i was wrong unless my I/V sounds worse than ADA4898 on 16-20 bit players .

Until that happens, I'm still into masking techniques as i consider them more valuable than cleaning everything .I don't really care how good a technology can be if it's not cheap. And i will never buy that Benchmark dac...In 10 years from now another guys will tell that their new toys are WAY better than a benchmark , the same as early CD players were advertised to be the best audio out here.
 

Attachments

  • curconv1.png
    curconv1.png
    70.8 KB · Views: 546
  • curconvb.png
    curconvb.png
    86.9 KB · Views: 542
  • curconvc.png
    curconvc.png
    83.5 KB · Views: 550
  • currenttovoltageabc.asc
    4.8 KB · Views: 44
I don't have copy...i have the actual third edition which i bought in late 2015.
I have hard copies (physical books) of both the 2nd and 3rd editions, since some material is only covered in one of the versions.

I'd guess that the dev boards sent by the dac manufacturers have very well selected op-amps for the job.

Well designed multilayer boards with power and ground planes can do wonders for opamps (and dacs in general). Passive component selection can matter too.
 
Account Closed
Joined 2010
I had a copy of the first edition in 2009, i think i still have it at home and i donated the second edition bought from second hand book shop, but i never had the time to read them all...
Anyway, i know a bit of pcb design , i'm not a master, but i know how to get the best out of an analog design.
I had to cancel the command for ada4898-2 as it wasn't stocked at mouser and launched a command for ada4898-1 and 4899-1 as the latter has about 300v/us slew rate and better footprint.

I also bought different audio op-amps (modern ones) and some of the lowest noise pnp bipolars to rebuild my version of I/V .I have no measuring instruments with me .I will only use my ears and i fully trust them. The circuit that can impress them first will win.
If anybody is willing to help me learn how to sim a true dac output in LTSpice please help me!
Once i finished with the i/v stage i'll probably buy a modern dac like the one described by this topic , add the winner i/v stage , a passive filter on the output of this stage and see if there's any improvement or revelation over tda1541 !
I'm truly curious, i'm not a fan of any technology whatsoever, i only go with the best illusion!
Given the fact that i only listen to some "good enough" headphones, until now the best technologies for me were only three:

1 a good cd player combined with a tape emulator or a good cassette player in dolby C(i have no working reel to reel unfortunately),

2 a good turntable with valve-transistors-hybrid preamps
3 a good cd player and a valve headphones amplifier)

Digital technology alone never impressed me enough maybe because the artists weren't good enough and i could hear them clearly as they really sound :)
Most probably i will never accept the best digital reproduction technology on the grounds that most of the music recorded today is not really pleasing to me.
 
Pachy

Getting back to original thread subject, it's worth going through the DDDAC
thread. There are a lot of tweaks to get the best out of 1794.

A NOS 192/24 DAC with the PCM1794 (and WaveIO USB input)

There's a lot to go through but a few things that were implemented:

a/ increasing OP current through current scaling resistor
b/ Using a CCS instead of that resistor for better noise immunity
c/ Increasing analog voltage supplies.

cheers

T
Yep, the thread really went south but the discussion was interesting so I don't mind at all.

I have done a few mods to the board after my last post...

1. Changed the LM317/337 regulators for the OPAMP supplies and the AVCC of the PCM1794 with high PSRR(-125dB), low output impedance discrete regulators.

2. Increased PCM1794 AVCC voltage from 5V to 6.3V. I like to adhere to the datasheet absolute maximum values, I know DDDAC has experimented with higher values but there is no point in doing this when you use an active I/V.

The sound became a little "colder" but more details "appeared", especially in the highs so I figured out that what I thought was "colder" actually was closer to the exact recording. Initially I felt like I liked the "warm" feeling more, but after listening for a few hours I found out I was wrong.

I think that there is not a lot more I can do with this board in terms of improvement, the sound quality has already exceeded my expectations.
 

Attachments

  • pcm1794_discrete_regs.jpg
    pcm1794_discrete_regs.jpg
    333.3 KB · Views: 551
Account Closed
Joined 2010
You may recall that properly dithered 16-bit recordings made from master taps contain information below the 16-bit level, it is just mixed in with some noise. Fortunately, human brains are good at extracting signals from noise, so there are details to recordings you have no idea even exist.
I don't know how capable is dithering at resurfacing music details from digital noise but...I'll see what I can do with these:
https://www.diyaudio.com/forums/dig...cd-players-enhancing-noise-6.html#post5713768
 
Well now that you’re satisfied with it, and think that it couldn’t get any better, maybe try a few more things;
Replace one of the values of the small capacitors in the output filter with a Seimens KS, or emz polystyrene.
Do some line conditioning on the incoming ac, a common mode choke, and a .1uf X rated capacitor works well.
Put snubber circuits onto the diodes, my dac uses a combination of a .01uf along with a .1uf in series with an 18 ohm resistor.
Replace the .1uf blue capacitors with some relcaps, polystyrene if you can, place one across the power pins of the op amps as well.
Ditch the Panasonic fc caps and use something not made for an smps.

Can try these one at a time and see how the changes present themselves to you. These have all been things that have improved the kits and consumer gear I have had.