Works if you keep truncating the results... This is the way how 7220 looses information: 30 multiplications of 16bit x 12bit are stored in a 16 bit acumulator. Obviously each result will suffer truncations.
Also, it is obvious that the DC offset will come off the final 16 bit resolution, because is added at the acumulator result. So the coeficients BEFORE that need to limit the total range of results with exactly that amount, to leave "room" for that DC (to avoid clipping). Obvious that none of th einput data is maintained.
PS: Other "bad" filters output 20-24 bit from 16 input data and feeds it to 20-24 bit DAC's for exactly same reason - keep the truncation errors to minimum. SM5846 uses 32 bit acumulator and 24 bit output. But that's offtopic
Last edited:
Hi,
This is true for each and every digital filter (including that in ESS Sabre reference DAC). The process is inherently lossy, as I pointed out above, repeatedly.
However the error remains within +/- 1/2LSB of the input signal.
As Philips do not disclose the precise nature of the accumator it may be quite large. As the SAA7220 Datasheet uses the phrase "the 16MSB from the accumulator (are output to the serial interface)" we can conclude it is larger than 16 Bit. It is entirely possible that 32 Bit are used, as others mentioned.
The use of only the 16 MSB's produces an error of +/-1/2LSB for each output sample.
But we have 4 samples for each input sample the combined integrated four samples together should reduce the error by 6dB and thus leave the input sample error to dominate.
The measurements of the best Philips SAA7220/TDA1541A based Players/DAC's do bear this out (try a Philips LHH1000 or Marantz CD-12/DA-12 combo with upgraded I/V conversion to illustrate this).
Ciao T
This is true for each and every digital filter (including that in ESS Sabre reference DAC). The process is inherently lossy, as I pointed out above, repeatedly.
However the error remains within +/- 1/2LSB of the input signal.
As Philips do not disclose the precise nature of the accumator it may be quite large. As the SAA7220 Datasheet uses the phrase "the 16MSB from the accumulator (are output to the serial interface)" we can conclude it is larger than 16 Bit. It is entirely possible that 32 Bit are used, as others mentioned.
The use of only the 16 MSB's produces an error of +/-1/2LSB for each output sample.
But we have 4 samples for each input sample the combined integrated four samples together should reduce the error by 6dB and thus leave the input sample error to dominate.
The measurements of the best Philips SAA7220/TDA1541A based Players/DAC's do bear this out (try a Philips LHH1000 or Marantz CD-12/DA-12 combo with upgraded I/V conversion to illustrate this).
Ciao T
Hi,
You are right, it is added to the accumulator, which basically sums up the weighted samples into the new sample. I treated Coefficients, Multiplier and Accumulator as in effect a single unit.
But we may disaggregate it further.
Philips does not specifically list the length of the accumulator, as only 16 the 16 MSB's are used. Anything below the 16 Bit LSB does not contribute to the output word, so the 16 Bit error of +/-0.5 LSB is retained, as no noiseshaping is included.
I also checked some other literature, it confirms no noise-shaping etc. in the SAA7220.
Though the SAA7030 that was matched with the TDA1540 did contain a first order noiseshaper (so Philips understood how to implement this technique), Philips decided not include it in the SAA7220/TDA1541 combination.
For a long time the SAA7220/TDA1541A combo was the mainstay of High End audio and only really dethroned by HDCD, the PDM100 Filter and the best 20Bit DAC's from AD and BB, introduced at a time when Philips had already abandoned the field to pursue low end cost efficiency and high volume with their bitstream offerings.
It was funny though that the first Philips CDP I owned was a massively overperforming low end offering sold to me by a discounter shop for the princely sum of 89 Pound Sterling and 99 pence. It used the TDA1545A (not really a patch on the TDA1541, but surprisingly good in it's own right) and what was essentially the core of the SAA7220 included in the servo processor.
This unit was modified by me with much improved power supplies, Current Feedback Op-Amp analogue stages and a "superclock" as well as stone plate attached to the flimsy case and for a long outperformed any comers.
This included the Maratz CD-67SE I bought unheard based on magazine reviews (I never did that again) and similar priced (400 pound class) players from the japanese majors, including ones from Denon with Alpha Processing, Pioneers with Legato Link, Technics players with a technology that rhymes on "mush" and others. It also sounded better than experimental DAC's I build at the time based around the heavily hyped Cirrus Logic DAC's...
Ciao T
You are right, it is added to the accumulator, which basically sums up the weighted samples into the new sample. I treated Coefficients, Multiplier and Accumulator as in effect a single unit.
But we may disaggregate it further.
Philips does not specifically list the length of the accumulator, as only 16 the 16 MSB's are used. Anything below the 16 Bit LSB does not contribute to the output word, so the 16 Bit error of +/-0.5 LSB is retained, as no noiseshaping is included.
I also checked some other literature, it confirms no noise-shaping etc. in the SAA7220.
Though the SAA7030 that was matched with the TDA1540 did contain a first order noiseshaper (so Philips understood how to implement this technique), Philips decided not include it in the SAA7220/TDA1541 combination.
For a long time the SAA7220/TDA1541A combo was the mainstay of High End audio and only really dethroned by HDCD, the PDM100 Filter and the best 20Bit DAC's from AD and BB, introduced at a time when Philips had already abandoned the field to pursue low end cost efficiency and high volume with their bitstream offerings.
It was funny though that the first Philips CDP I owned was a massively overperforming low end offering sold to me by a discounter shop for the princely sum of 89 Pound Sterling and 99 pence. It used the TDA1545A (not really a patch on the TDA1541, but surprisingly good in it's own right) and what was essentially the core of the SAA7220 included in the servo processor.
This unit was modified by me with much improved power supplies, Current Feedback Op-Amp analogue stages and a "superclock" as well as stone plate attached to the flimsy case and for a long outperformed any comers.
This included the Maratz CD-67SE I bought unheard based on magazine reviews (I never did that again) and similar priced (400 pound class) players from the japanese majors, including ones from Denon with Alpha Processing, Pioneers with Legato Link, Technics players with a technology that rhymes on "mush" and others. It also sounded better than experimental DAC's I build at the time based around the heavily hyped Cirrus Logic DAC's...
Ciao T
Hi,
Not only that, it was also inserted after I posted my reply. Bad form old chap, bad form.
Ciao T
PS: Other "bad" filters output 20-24 bit from 16 input data and feeds it to 20-24 bit DAC's for exactly same reason - keep the truncation errors to minimum. SM5846 uses 32 bit acumulator and 24 bit output. But that's offtopic
Last edited by SoNic_real_one; Today at 11:21 AM.
Not only that, it was also inserted after I posted my reply. Bad form old chap, bad form.
Ciao T
I would say so, too.
AYA DAC is not in the production any more. Those kits were all based on AD844 that exhibit close to 1% distortion when used the way Pedja had used them - something I find unlistenable. Later, his finished products incorporated much better implementations of all stages, not only analog ones -> that no doubt sound exceptional. Maybe you can get one of his finished DAC's and be done with it?
I've one of his AYA II DACs with the OPA861 I/V stage and I'm satisfied. Have tried using passive I/V at 33R from the 1541A (same DAC) then tube stages of all sorts, but keep coming back to what he had done. One of the nicer ones was WE396A CCS loaded and parafed into an S&B 102mk3 (as an autoformer), but came back to what Pedja had.
If I was looking, I'd buy his commercial offering (TDA1541A), I doubt you'd look back.. he knows what he's doing, and for the money.. you really cant go wrong.
AYA DAC is not in the production any more. Those kits were all based on AD844 that exhibit close to 1% distortion when used the way Pedja had used them - something I find unlistenable. Later, his finished products incorporated much better implementations of all stages, not only analog ones -> that no doubt sound exceptional. Maybe you can get one of his finished DAC's and be done with it?
I've one of his AYA II DACs with the OPA861 I/V stage and I'm satisfied. Have tried using passive I/V at 33R from the 1541A (same DAC) then tube stages of all sorts, but keep coming back to what he had done. One of the nicer ones was WE396A CCS loaded and parafed into an S&B 102mk3 (as an autoformer), but came back to what Pedja had.
If I was looking, I'd buy his commercial offering (TDA1541A), I doubt you'd look back.. he knows what he's doing, and for the money.. you really cant go wrong.
Thorsten, man I 've been around the block with tube stages, actually your 10R cathode biased ECC88/6922 stage was the first I tried, choke and CCS loaded. Yes!, it was good, but bettered by AYA II I/V. Easily.
I've just read 29 pages worth and would like to ask for a verbal (just quick) description on your 'final' tube OP stage. 6072A, if you wouldnt mind. I know its a well regarded tube, used by high brand names, but almost exclusively as the main voltage gain stage and then its buffered. According to the sheet, gain is around 40, which is nice, but Rp is like 25k. Are you using a step down xfmr, or mu output from a CCS load to reduce Rp?.
Sincerely..
I've just read 29 pages worth and would like to ask for a verbal (just quick) description on your 'final' tube OP stage. 6072A, if you wouldnt mind. I know its a well regarded tube, used by high brand names, but almost exclusively as the main voltage gain stage and then its buffered. According to the sheet, gain is around 40, which is nice, but Rp is like 25k. Are you using a step down xfmr, or mu output from a CCS load to reduce Rp?.
Sincerely..
Basically same as the 6922 one.
Btw, that stage is biased by the TDA1541 offset and on the Aya you will have to remove the offset current injection.
For the 6072 I use a Fet based CCS with the output taken from the "Mu-Follower" output of the CCS...
I suspect you just happen to like a slightly different sound, so you don't so much care for Tubes...
Ciao T
This is a good read that I think pertains to a tube dac stage:
http://www.diyaudio.com/forums/diyaudio-com-articles/163570-his-masters-noise-thoroughly-modern-tube-phono-preamp.html
What are thoughts on transformers in the output stage. My right brain is looking at all the hub-hub of galvanic isolating the USB from the DAC, but I have often wondered if it is more important to use quality power from usb foward and just have the digital section of the DAC share ground with the computer. Provide the galvanic isolation in the analog section.
If we use a step up transformer as done in the article for the phono stage, the issue is impedance. Take a typycal low-level 1:10 step up transformer. Primary is generaly very low DCR, so an I/V resistor is only going to lower the I to V conversion further. But this primary is also low inductance, not suitable inductance for a DAC i-out of 1k to 2k ohm. I just don't know how you really get a step transformer to help I/V, seems too many issues and I have had little luck.
But using a 12 ohm I/V resistor with a D3A common cathode, gives 3.46/2*.707= 1.2Vrms output. Of course high quality noise free B+ and F+ are critical. But one would think we could extract close to -~-90db THD and SNR. We could stop here and use the computer to boprovide the compensation and the output impedance is a respectable 2.5k.
But we may want some filtering and of course the compensation in the analog stage, which wouldn't be prudent at a 50mVrms level, so it could be put after the D3A, again analogous to the phono preamp. But now we need an output tube that is quiet, has a gain of less 3. Isn't there an indirect heated version of the 2A3? Or ofcourse we make the second stage a 6922/5687/6n6/etc tube with an pre-amp output transformer, maybe even a DHT for the adventourous.
I did see the comments about leaving transformers out of the signal path in a DAC, but just seems they can be an advantage. You really don't see opamp I/V with significantly less distortion than tubes/mofsets in the analog stage and surely opamps put loads of PS capacitance in the signal path.
Alternatively there is the Zen with parralel 2SK372 jfets which can keep the input impedance below 10 ohms and can be transformer coupled, but simulations have showed much worse THD/IMD than tubes, I haven't simulated the Cen/Sen.
http://www.diyaudio.com/forums/diyaudio-com-articles/163570-his-masters-noise-thoroughly-modern-tube-phono-preamp.html
What are thoughts on transformers in the output stage. My right brain is looking at all the hub-hub of galvanic isolating the USB from the DAC, but I have often wondered if it is more important to use quality power from usb foward and just have the digital section of the DAC share ground with the computer. Provide the galvanic isolation in the analog section.
If we use a step up transformer as done in the article for the phono stage, the issue is impedance. Take a typycal low-level 1:10 step up transformer. Primary is generaly very low DCR, so an I/V resistor is only going to lower the I to V conversion further. But this primary is also low inductance, not suitable inductance for a DAC i-out of 1k to 2k ohm. I just don't know how you really get a step transformer to help I/V, seems too many issues and I have had little luck.
But using a 12 ohm I/V resistor with a D3A common cathode, gives 3.46/2*.707= 1.2Vrms output. Of course high quality noise free B+ and F+ are critical. But one would think we could extract close to -~-90db THD and SNR. We could stop here and use the computer to boprovide the compensation and the output impedance is a respectable 2.5k.
But we may want some filtering and of course the compensation in the analog stage, which wouldn't be prudent at a 50mVrms level, so it could be put after the D3A, again analogous to the phono preamp. But now we need an output tube that is quiet, has a gain of less 3. Isn't there an indirect heated version of the 2A3? Or ofcourse we make the second stage a 6922/5687/6n6/etc tube with an pre-amp output transformer, maybe even a DHT for the adventourous.
I did see the comments about leaving transformers out of the signal path in a DAC, but just seems they can be an advantage. You really don't see opamp I/V with significantly less distortion than tubes/mofsets in the analog stage and surely opamps put loads of PS capacitance in the signal path.
Alternatively there is the Zen with parralel 2SK372 jfets which can keep the input impedance below 10 ohms and can be transformer coupled, but simulations have showed much worse THD/IMD than tubes, I haven't simulated the Cen/Sen.
Hi,
The circuit shown is mainly taken from my friend Steven Robinson's design (look at the earlier versions):
http://www.izzy-wizzy.com/audio/preampnew.html
Some of the design choices are questionable, especially from a noise viewpoint.
As the TDA1541 has -2mA offset at digital silence, this can be used to bias the output stage. As remarked before, there is no reason to slavishly adhere to the 25mV compliance spec of the datasheet. It can be exceeded significantly.
Just take a TDA1541A DAC, any decent spectrum analyser (can even be a PC, RMAA and a 300 Euro EMU 1616m) and test for yourself how much resistance is really tolerable.
Ciao T
The circuit shown is mainly taken from my friend Steven Robinson's design (look at the earlier versions):
http://www.izzy-wizzy.com/audio/preampnew.html
Some of the design choices are questionable, especially from a noise viewpoint.
As the TDA1541 has -2mA offset at digital silence, this can be used to bias the output stage. As remarked before, there is no reason to slavishly adhere to the 25mV compliance spec of the datasheet. It can be exceeded significantly.
Just take a TDA1541A DAC, any decent spectrum analyser (can even be a PC, RMAA and a 300 Euro EMU 1616m) and test for yourself how much resistance is really tolerable.
Ciao T
Maybe that's because you didn't look around. ESS published data shows I/V based on OpAmp with measured THD+N at -114..116dB.
Of course, that does not come from a TDA1541 that cannot dream of something like that. It just proves that OpAmps are not the weakest link and you are tring to "fix" something that is not broken.
Last edited:
Thanks, I can try it..
Hard to say why my preference for the OPA861 as he has it.. the closest I can get in reasoning with it is that I can null the voltage at the output of the chip to zero volts and direct couple to the S&B102. If I use the tube stage I need to either cap couple it out (or use yet another transformer).. and I always seem to hear that cap. Anyway, I'd try the 6072 for kicks.. had thought about using a 6SL7 CCS loaded the same way and smaller IV resistor. Actually, perhaps the lower voltage present at the output of the 1541A with the OPA cct has something to do with the subjective preference I have for it over 33R IV then tube. Maybe the 6SL7 really is worth a try here.. output impedance is certainly no longer an issue with a mu-out from a CCS and with higher gain IV resistor can be alot lower, you know the voltage compliance thing as per the 1541A data sheet.
I concur that a correctly selected opamp is probably not the weakest link. The DACs are - their dynamic performance sucks and that's a good reason to prefer the slowest sample rate possible - i.e. NOS.
Hi,
What are you using as Cap? Try Tinfoil & PP or FEP and if you can, apply wooden or plastic sleeves (or copper) to them and epoxy the capacitors in place. Also, power supplies play a major role in tube circuit sonics.
I am running at only the bias from the TDA1541 and with IIRC 33Ohm I/V, current I think around 4mA, Anode voltage around 100V, around 10V across the CCS.
The CCS is a "Ring of 2" with a low capacitance IRF Mosfet...
PSU is a combo BYV26 & Schottky rectifier, 100uF/20H/100uF/1K/100uF/1K/100uF and 1K/66uF Film per channel.
Ciao T
What are you using as Cap? Try Tinfoil & PP or FEP and if you can, apply wooden or plastic sleeves (or copper) to them and epoxy the capacitors in place. Also, power supplies play a major role in tube circuit sonics.
I am running at only the bias from the TDA1541 and with IIRC 33Ohm I/V, current I think around 4mA, Anode voltage around 100V, around 10V across the CCS.
The CCS is a "Ring of 2" with a low capacitance IRF Mosfet...
PSU is a combo BYV26 & Schottky rectifier, 100uF/20H/100uF/1K/100uF/1K/100uF and 1K/66uF Film per channel.
Ciao T
Isn't the above in contraddiction to this one:
I'd go any time for a solution like Pedja has chosen (he runs the OPA860/861 open loop AFAIK), Pass D1, Jockos "easy to build I/V", all common Base/Gate, open loop. SEN/CEN might be used if you parallel enough FET's to come below 10 Ohms input Z.
You can easily make Pedja's I/V even better - just use a pair of paraleled AD844s per channel. You get lower input impedance, and higher current available at TZ pin.
There is a scheme which got many positive feedbacks over russian diy community, with passive LP filter and BUF03 on the output.
I'll make 3 versions of it, with various buffer options (LME49600, LH0033, LH0063) and Nazar's "electronic capacitors".
The scheme is from SergioT, who referenced it to Pedja.
There is a scheme which got many positive feedbacks over russian diy community, with passive LP filter and BUF03 on the output.
I'll make 3 versions of it, with various buffer options (LME49600, LH0033, LH0063) and Nazar's "electronic capacitors".
The scheme is from SergioT, who referenced it to Pedja.
Hi,
No, I was referring to exceeding the voltage established by measuring, not the datasheet one. Though going over a little does not kill sound quality completely, there is quite a bit of leeway.
I personally would select the "CEN" as the one I like conceptually best. Using 1pcs each of 2SK170BL/2SJ74BL at Idss gives around 12 Ohm Zin which if the output offset is nulled gives +/-25mV... In fact, even double the voltage will not be a disaster.
In practice I listen to passive I/V and Tubes though.
Ciao T
No, I was referring to exceeding the voltage established by measuring, not the datasheet one. Though going over a little does not kill sound quality completely, there is quite a bit of leeway.
I personally would select the "CEN" as the one I like conceptually best. Using 1pcs each of 2SK170BL/2SJ74BL at Idss gives around 12 Ohm Zin which if the output offset is nulled gives +/-25mV... In fact, even double the voltage will not be a disaster.
In practice I listen to passive I/V and Tubes though.
Ciao T
Hi,
The OPA860 is an uncommitted diamond transistor. Application is obvious, just read the datasheet, with your extensive experience with the TDA1541 I cannot see any issues.
Ciao T
The OPA860 is an uncommitted diamond transistor. Application is obvious, just read the datasheet, with your extensive experience with the TDA1541 I cannot see any issues.
Ciao T
Interesting circuit. I can accept C1 and C2 only if the input impedance is much lower than 1 ohm... otherwise I would not use them there.
I was referring to complete OPA861 circuit, with the buffer, that Pedja used in his final DAC kits just before everything was removed from his site.... If anyone has saved the circuit it would be interesting to see it and compare it with other solutions....
Boky
C1/C2 should be placed on DAC's side, to take down the HF noise and kinda limit the bandwidth.
2.2n value is pretty high tho, but it may vary on actual circuit. Maybe some 0805 np0 of 10-1000pF will cure it better, if soldered directly on ICs pins.
Still, you need to null the 1541's output with current source.
2.2n value is pretty high tho, but it may vary on actual circuit. Maybe some 0805 np0 of 10-1000pF will cure it better, if soldered directly on ICs pins.
Still, you need to null the 1541's output with current source.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.