I would like to build(I will use assitance form local experts, as my digital knowledge are very limited) a dac with 8x (or 16x) pcm63p-y chips . fully balanced (1xpcm63 for phase, minimum 4 psc required for stereo xlr output) Money is not issue(have the chips and still many black gate N left) on the design , so feel free ofer compromiseless sugestions. dac will be fed from usb/spdif/i2s converter.
however lot of questions- even before deciding haw to go-
1) what logic chips needed to conect i2s to pcm63(without digital filer and receiver) if not possible then-
2)is it possible to use digital receiver wm8804(or any similar quality chip_ straight to pcm63(without digital filter) if yes , what parts is needed.
3)are there some better.new filters than sm5842APT, which can be used with pcm63.
4) is it possible to run pcm63 straight to power amp input- a) without any circuit(just IV resistor) b) trought transformer at this stage I am not considering discrete IV(like pass labs d1, or zen IV which is fine with 4xpcm63)) with 16x pcm. I belive 16x pcm63p will have enought current???
aditional questions -how to drive 16xpcm63pk(several digital filter,receivers or logic chips needed? I believe it cant be done with just ordinary schematic for 2 or 4x pcm63)
maybe some pcm63p implementation ideas?
will wait answers to what direction its best to go.
P.S. yes I have moder DAC's (with ess9018,1794 ) but always dreamed to have OLD dac with pcm63
I also have a old style dac (with wm8804>>sm5842apt>>4x pcm63p-k>>jfet output stage) which I like the most, and would want do better, or to say best.
thanks again for reading, and hoping to get some advices..
however lot of questions- even before deciding haw to go-
1) what logic chips needed to conect i2s to pcm63(without digital filer and receiver) if not possible then-
2)is it possible to use digital receiver wm8804(or any similar quality chip_ straight to pcm63(without digital filter) if yes , what parts is needed.
3)are there some better.new filters than sm5842APT, which can be used with pcm63.
4) is it possible to run pcm63 straight to power amp input- a) without any circuit(just IV resistor) b) trought transformer at this stage I am not considering discrete IV(like pass labs d1, or zen IV which is fine with 4xpcm63)) with 16x pcm. I belive 16x pcm63p will have enought current???
aditional questions -how to drive 16xpcm63pk(several digital filter,receivers or logic chips needed? I believe it cant be done with just ordinary schematic for 2 or 4x pcm63)
maybe some pcm63p implementation ideas?
will wait answers to what direction its best to go.
P.S. yes I have moder DAC's (with ess9018,1794 ) but always dreamed to have OLD dac with pcm63
I also have a old style dac (with wm8804>>sm5842apt>>4x pcm63p-k>>jfet output stage) which I like the most, and would want do better, or to say best.
thanks again for reading, and hoping to get some advices..
Search for threads by iancanada for what you need to get the bits into the DAC chips. Sounds like a very ambitious project. Good luck.
Thanks very much for the input, now I have some basic start. In my pcm63dac i had cs8412, then 8424 (i did not compare side by side but it seems 8412 was not worse, maybe slight better) on the some dac its designer used dir9001, and after some years switched to wm8804, which he thinks(and measures) better than dir9001.
OK so for now starting point - sm5842apt or DFI1704 filter(pcm1794a sound wonderful in NOS mode- less conguested, more efortless comparing with OS/ASCR) question again- how much gain of amp is nessesary(and what input impedance) to run dac chips directly?
as for accuphase- it can be bought cheaply these days(one is sold today for 3000usd) but its nightmare for mods (no space) so better start from scrach. I have listened to dc-91, and I would not say its was anything special. With today more moder chips and diferent IV and biger enclosure size better sound can be achieved from 16xpcm63, not speaking about pcm63p-y grade
Last edited:
I believe I know Ian (FIFO maker), I read this forum for decades, and are amazed of digital knowledge of some members. Would put many comercial engineers to shade.
Without help of those members this project will not happen or likely worse will happen but ill be only working (I know that good sound is not a chip, but implementation) so far i can ask my engineer(he is into amps, not digital) to draw shematic, make pcb, thats all. I very rarely ask for help, and always try to help for speaker builders, so hope somebody will drop me genius ideas..
question nr3 if raw dac chip can not be put into 35-40db gain amp (I can build amp wth 2-100kohms impedance) is there point use 16xpcm63(lot of place on pcb) in another words- will i gain something using transformer output 8x vs 16psc?
the most thing i am concerned is digital stage (analog can be be modified later)
I'd run NOS as that's lower overall noise, lower glitching. Passive I/V followed by passive filtering. Using 16 paralleled PCM63s should in theory average out the bit weight errors - you could also dispense with analog domain volume control by gradually reducing the contribution of the DACs one by one using DSP, rather than paralleling them all.
Sounds like a fun project!
Sounds like a fun project!
In my view, NOS adds more problems than it allegedly solves. It really necessitates the use of an aggressive, near-sonic LPF, which brings back all of the horrid phase problems of early digital. There's just no way to do it, IMO, that doesn't do more damage than a good 8x OS that has no need for a LPF in practice.
The reduction of dac numbers to reduce volume will work, but means restricted dynamic impact & note decay at lower volumes, and you would only have 8 volume steps in this case, since each signal phase would have only 8 dac chips. WAY too coarse for practical use, IMO.
If I were going to do purely passive between dac & amp, in this situation, I would use Cinemag transformer out, feeding fixed full scale output level through the dac-to-amp cable, then have a stereo Alps RK27 pot or DACT stereo stepped attenuator configured as balanced volume control for each channel directly inside the amp chassis, as close as possible to the input stage.
If you had an amp with a gain of something like 80dB(just guessing), you could theoretically have the current out of each set of 8 dacs connected straight to the output jacks, then a 10ohm wirewound dual pot configured as balanced volume control at input of amp. That would be really interesting to hear.
The reduction of dac numbers to reduce volume will work, but means restricted dynamic impact & note decay at lower volumes, and you would only have 8 volume steps in this case, since each signal phase would have only 8 dac chips. WAY too coarse for practical use, IMO.
If I were going to do purely passive between dac & amp, in this situation, I would use Cinemag transformer out, feeding fixed full scale output level through the dac-to-amp cable, then have a stereo Alps RK27 pot or DACT stereo stepped attenuator configured as balanced volume control for each channel directly inside the amp chassis, as close as possible to the input stage.
If you had an amp with a gain of something like 80dB(just guessing), you could theoretically have the current out of each set of 8 dacs connected straight to the output jacks, then a 10ohm wirewound dual pot configured as balanced volume control at input of amp. That would be really interesting to hear.
I've built several aggressive LPFs and agree they're really necessary in NOS, however 'horrid phase problems' have not been audible with any of the filters I've built. The addition of the LPF has been audibly beneficial, improving the noise floor.
On the other hand, using OS simply increases the noise floor (glitching primarily, digital crosstalk secondarily).
As for how many digital volume steps - depends on the step resolution. Its not a universal panacea I agree and should be weighed against any issues from a proposed analog volume control.
On the other hand, using OS simply increases the noise floor (glitching primarily, digital crosstalk secondarily).
As for how many digital volume steps - depends on the step resolution. Its not a universal panacea I agree and should be weighed against any issues from a proposed analog volume control.
More to add.
In my view, there can be no such thing as an NOS bitstream dac, regarding your PCM1794 comment. The d-s modulator is an oversampler itself, with an OS d.f. merely adding more OS. If you look at a sinewave output from an "NOS" PCM1794 versus the same from a ladder dac, it's pretty obvious one has little relationship to the other. Also, the d.f. within the PCM1794/etc. chips seems known to be markedly inferior to good d.f. chips like the DF1704 and SM5847.
I certainly agree with you on the sound of a stock or "conventionally upgraded" Accuphase DC-91. For god knows what reason, other than it was common practice at the time, each of the 32 pcm63 chips has it's own feedback i/v opamp, then into a common buffer, then into a HORRIBLE GIC filter, then output buffer and balanced driver amp. With single common passive i/v resistor per 16 dacs per channel, then my no-feedback voltage gain + line driver stages, plus way larger caps around each of the dacs(over 500 nice Nichicon caps I had to replace with larger nice Nichicon caps, exhausting), the resulting sound is absolutely spectacular, without even changing the ancient YM3436 input receiver or d/f arrangement.
In my view, there can be no such thing as an NOS bitstream dac, regarding your PCM1794 comment. The d-s modulator is an oversampler itself, with an OS d.f. merely adding more OS. If you look at a sinewave output from an "NOS" PCM1794 versus the same from a ladder dac, it's pretty obvious one has little relationship to the other. Also, the d.f. within the PCM1794/etc. chips seems known to be markedly inferior to good d.f. chips like the DF1704 and SM5847.
I certainly agree with you on the sound of a stock or "conventionally upgraded" Accuphase DC-91. For god knows what reason, other than it was common practice at the time, each of the 32 pcm63 chips has it's own feedback i/v opamp, then into a common buffer, then into a HORRIBLE GIC filter, then output buffer and balanced driver amp. With single common passive i/v resistor per 16 dacs per channel, then my no-feedback voltage gain + line driver stages, plus way larger caps around each of the dacs(over 500 nice Nichicon caps I had to replace with larger nice Nichicon caps, exhausting), the resulting sound is absolutely spectacular, without even changing the ancient YM3436 input receiver or d/f arrangement.
I can't grasp why you assert that adding an OS chip increases the noise floor. If it does, then it's being done completely wrong. My results with various ladder-dac-based units with OS have proven conclusively to me and my clients & friends that properly done OS has absolutely zero noise penalty in practice, unless you have loop feedback translating/exaggerating the very high frequency sampling noise into audible haze.
Running an R2R DAC at 8X OS means its spending a larger proportion of its time settling to its final value. This is quite independent of the nature/topology of the output stage.
Well, except for cruddy old TDA1543 dacs, I can't see settling time being a problem, and certainly can't hear it being a problem.
Not true. Settling time is proportional to the step size. 8X OS means the average step is eight times smaller than NOS and the total time spent settling is about the same. If you analyse the content of CD tracks you will see the RMS step size is rather small.The RMS step size of Hi Rez files is even smaller. Settling time is really not an issue. The real crime is delta-sigma, which never settles and is not even monotonic!
What you propose is similar to my current project. Here are some things to consider:
If your source is USB you don't need S/PDIF or I2S. Feed the PCM63 directly from the USB stream at up to 768K. Do the resampling in the PC. It's better then any DF chip. 32 DACs per channel fits very comfortably with off-the-shelf digital logic. Fully balanced is not always the best choice. Make both balanced and parallel operation an option. With 8 or more DACs per channel adding linear interpolation as an option is a no-brainer.
Last edited:
Thanks again everybody for briliant inputs, we are mooving forward. I have no technical knowledge about NOS PCM1794 dac I just refered to this DAC- http://www.fetaudio.com/wp-content/uploads/2012/04/FDA-1B-DAC-Manual_r2a.pdf
Now about ability of direct pcm to RCA. About what exact driving ability and values we are talking? For example with 400w@4ohm amplifer (conected to 3ohm speakers), which have 26-28db gain and 30kohm input impedance- to how many wats it can be driver from balanced pcm63 output(4pcm63pl paraleled per phase) or single eneded(8pm63p per chanel)?? for comaprission most balanced dac with 6V output can drive 26-28db amplifier to its max( or nearly to max, as it really depends on record levels) swing? I belive accuphase dc-91 have 8pcm63 per channnle not 16. I would like to stay with 16 pcm63p-y chips per both channels( I have only limited numbers of those, and do not trust on internet offfers of those chips authenicity)
I can build special amp with 80db gain, but would not want to use this dac with only this amp. So after i get exact numbers of output abilities in balanced or single ended mode , we will decide if it possible to drive example amp(26db/30kohm) at least for 50wats If answer is SO SO I can do selectable output direct or via transformer, iF answer is strictly NO, only ransformer option will be selected. In this case again I need know which type transformer (ratio) I need to achive 6V output to what input impedance.
I believe a very good dac can be made with just one or two chips per chanel with ordinary IV stage, but my intentions of using more chips opens new(unusualy) possibility (unusual implementations sometimes leads to unusual results, and I can take a risk considering this desisions is based on some science ) of IV stage. I have many usual dac so this one is for diferent purpose.
BTW considering my sonic targets- I dont care too much about high resolution records, or auddiophile etched details with spitting midrange in your face. I value soundstage depth,height and width (for the same reason i have very big speaker, which tweeter midrange sits 1.7meter from floor) even if sacrife exact pin point focus, and very fast/hughe macro dynamyc (listen for windband orcehstra mainly , no jazz gimmies which need details, fine tone,and very fine microdynamyc) The bass would nessesary be nothing but weighty(this is why i liek pcm chips- in most of it clever implementations it does it like no oneother chip)
However i DO NOT like ROUNDED type of sound(PCM63P is already have tendensy to sound slight less sharp edge comparing to my taste) . Friend of mien have Audio Note Fifth element DAC (I believe silver transformers everywhere and NOS) it it still have that rounded type of sound of DAC5 signature , and especialy lower Audio Note NOS DAC's. SO I am very intrigued by Tam LIn
however
another option is to acquire extra 16psc pcm63p-y chips (they would be extra 2000usd or so) , buty its not easy task to get them genue. and project will stop for another half year or so.
Now about ability of direct pcm to RCA. About what exact driving ability and values we are talking? For example with 400w@4ohm amplifer (conected to 3ohm speakers), which have 26-28db gain and 30kohm input impedance- to how many wats it can be driver from balanced pcm63 output(4pcm63pl paraleled per phase) or single eneded(8pm63p per chanel)?? for comaprission most balanced dac with 6V output can drive 26-28db amplifier to its max( or nearly to max, as it really depends on record levels) swing? I belive accuphase dc-91 have 8pcm63 per channnle not 16. I would like to stay with 16 pcm63p-y chips per both channels( I have only limited numbers of those, and do not trust on internet offfers of those chips authenicity)
I can build special amp with 80db gain, but would not want to use this dac with only this amp. So after i get exact numbers of output abilities in balanced or single ended mode , we will decide if it possible to drive example amp(26db/30kohm) at least for 50wats If answer is SO SO I can do selectable output direct or via transformer, iF answer is strictly NO, only ransformer option will be selected. In this case again I need know which type transformer (ratio) I need to achive 6V output to what input impedance.
I believe a very good dac can be made with just one or two chips per chanel with ordinary IV stage, but my intentions of using more chips opens new(unusualy) possibility (unusual implementations sometimes leads to unusual results, and I can take a risk considering this desisions is based on some science ) of IV stage. I have many usual dac so this one is for diferent purpose.
BTW considering my sonic targets- I dont care too much about high resolution records, or auddiophile etched details with spitting midrange in your face. I value soundstage depth,height and width (for the same reason i have very big speaker, which tweeter midrange sits 1.7meter from floor) even if sacrife exact pin point focus, and very fast/hughe macro dynamyc (listen for windband orcehstra mainly , no jazz gimmies which need details, fine tone,and very fine microdynamyc) The bass would nessesary be nothing but weighty(this is why i liek pcm chips- in most of it clever implementations it does it like no oneother chip)
However i DO NOT like ROUNDED type of sound(PCM63P is already have tendensy to sound slight less sharp edge comparing to my taste) . Friend of mien have Audio Note Fifth element DAC (I believe silver transformers everywhere and NOS) it it still have that rounded type of sound of DAC5 signature , and especialy lower Audio Note NOS DAC's. SO I am very intrigued by Tam LIn
Is it actualy posible? and HOW? never hear about this. PC(with Sotm usb card.linear suplies and windows server optimised software ) will be my only source.
however
if more chips is nessesary problems is that my dac chips qty is somewhat limited. Of cpurse I could always swith from 8 or 16 pcm63p-y per 2 chanels to 32psc of pcm63p-k chips per 2 channels. but i believe pcm63p-y chips is better than p-k grade.
another option is to acquire extra 16psc pcm63p-y chips (they would be extra 2000usd or so) , buty its not easy task to get them genue. and project will stop for another half year or so.
I do wonder why you choose PCM63 when a 16bit chip looks to be suitable for your needs. Try an array of TDA1387 and be sure to use a passive shunt supply to ensure best possible bass depth. Their price is a fraction of the PCM63's so you'll have plenty of money over for picking the very best caps to use in your passive shunt.
Hmm. I am not familar with this chip , and upon a quick search on over 4000psc variuos manufacturers dac's I wasnt able to find any comercial dac using it. Is it better(in general) than PCM63? As for price, its not the price factor that stops me from using 32psc pcm63p-y, its availability factor . As I told price is not issue (I am willing to spend 10 000USD for this DAC parts,(excluding case and transformers) and believe it be enought for most wildest and crazy engineering ideas. 10 000USD into DIY converts to something 200 000-300 000USD in comercial dac's and its already zone where DIY product can be created not because comercial unit can not be afford, but because nobody in market design such designs because of cost. If needed i can add more funds, no problem here. what i 'm lacking is knowledge what with those resourses can be achieved. whan i will have theoretical directions I will ask my engineer to fine tune them.
The problem is that its only one dac will be made, so my question how to involve some god gifted engineers here to help for one forum member(when work included in projects like FIFO, Bufallo have more wide spread usage) when more logical is to design something for the masses.
as for the moment lets wait for Taip Lin input- I realy excited about skipping i2s and conect usb to pcm63
Last edited:
http://www.diyaudio.com/forums/digi...56-pcm63-others-nos-mode-pc-oversampling.html
I considered using this for my project but it only goes to 384K and the XMOS approach is stupid overkill - who needs a 32-bit processor running at 500 MIPS to assemble a pair of 1.5 MB/s data streams? I'm making my own USB interface.
As I've not listened to PCM63 critically against TDA1387 I can't answer your question. However I'm of the belief that architecturally, the PCM63 is beaten by the Philips design because PCM63 is R2R internally whereas the TDA is of segmented current source design, hence the latter is subject to lower glitch. Of course the TDA is limited to only 16bits - the PCM63 is 20bits.
As you're willing to put up so much cash towards this project I'm curious why you focus on the DAC as if its a separate component, rather than taking a system-wide approach. What system is your DAC going to slot into - I wonder if the overall result could be further optimized by having (say) a line-level or digital crossover?
As regards going from USB direct into the PCM63, I can't see how that's possible without a fairly powerful CPU (or FPGA with soft CPU) in between the two. LPC4320 would do the job I think as it has both a high-speed USB and a generic serial interface on-chip, but programming it might be beyond your current skill set.
As you're willing to put up so much cash towards this project I'm curious why you focus on the DAC as if its a separate component, rather than taking a system-wide approach. What system is your DAC going to slot into - I wonder if the overall result could be further optimized by having (say) a line-level or digital crossover?
As regards going from USB direct into the PCM63, I can't see how that's possible without a fairly powerful CPU (or FPGA with soft CPU) in between the two. LPC4320 would do the job I think as it has both a high-speed USB and a generic serial interface on-chip, but programming it might be beyond your current skill set.