Well, the input signal is still 44.1k. Each sample is replaced by 4 samples with interpolation between samples by the digital filter, however averaging brings us back to 44.1kHz/16 Bit in analogue terms.
Then do that. With 1024X MCK we need a division by 256. My design includes 4 Flip Flops (74F74) to divide by 16.
We need an equal divide by 16 to use MCK and end up with 176.4/192kHz Fdem.
Or use a different divider IC.
Thor
off topic, true the PM75 is a little weak to manage the bass on a normal low hifi loudspeaker (<90 db/2.83V) with average ohmic curve (say not less than 4 to 3 ohms in the lows) ? Ahaha, most of the second hands PM75 should have their s1 removed for a basic one !
WOnder how it sounds recaped if needed. There was some good made dev (yamaha (I still like the meaat sound of my CX2), Marantz, Philips (ah the lowed blue MKT), etc back then ) and the choice of the passive parts not by chance (sometimes Elna cerafine, BG, good old Nich and Pany !)
WOnder how it sounds recaped if needed. There was some good made dev (yamaha (I still like the meaat sound of my CX2), Marantz, Philips (ah the lowed blue MKT), etc back then ) and the choice of the passive parts not by chance (sometimes Elna cerafine, BG, good old Nich and Pany !)
No idea. I use Technics SB-E100, 6 Ohm / 95dB/2.83V.
Yes, I got various high grade parts for it.
Thor
Rigth or not : sometime old caps which a certain "tonality" (it is a shortcuts we know all caps have no sounds but tech chatacteristics that differ from brand to brand) are not so bad with more leakage and more ESR according the area where it's used ! Leakage can be a problem though !
I still love my blue Nich SX caps (80s' and 90s')more than the VX that was said to be better (but not) ! I wish I had more of the old blue Pany !
okay, let's go back to the topic that is electronic shematic and not piggs lipstick to make it prettier ! 😊
I still love my blue Nich SX caps (80s' and 90s')more than the VX that was said to be better (but not) ! I wish I had more of the old blue Pany !
okay, let's go back to the topic that is electronic shematic and not piggs lipstick to make it prettier ! 😊
hi thor, I'm using DEM Grunding without filter (NOS) and in SIM mode. Also my caps are: 1x 4.7uF + 1x 1uF + 5x 0.47uF. I must say that it works without problems but I would like to verify that it is optimally sized. How could I do it?
Thanks
Antonio
It is not addressed to me, but here is my view on this:
A filter is a filter. Philips designed the DEM oscillator to operate at about 200 kHz, with filter (they call it decoupling) capacitors 100 nF. All capacitors are the same.
I analyzed in the other thread that the source impedance of the DEM pins is in the order of 10 Mohm. Also the ripple amplitude is about the same at all pins. The ripple frequency is fDEM/2 or fDEM/4 (source: Thor & John, IIRC). Consequently there is no reason to use higher capacitor at the MSB pins, because the goal is to filter the ripple amplitude below < 1/2 LSB, regardless of which bit current is being filtered. Note the DEM system works only at bits # 9 to #15, the upper 7 bits. The unfiltered ripple would cause bit errors (and artefacts in the audio band), and our goal is to push this below 1/2 LSB. There is also a static bit current error at the upper 7 bits, I deal with this in the thread referred before. At the factory the chips were being selected for static bit errors, resulting in R1, no mark, S1 grades.
If you use NOS mode and the externally forced DEM frequency is equal to WS that is 44.1 kHz, you can scale up the capacitors to 4x value, that is 4 x 100 nF. 470 nF looks a reasonable choice, for all 7 DEM filter pins. Nothing is against using higher value capacitors if you like, but you won't gain anything.
Back in this thread, John @ecdesigns experimented with 50 Hz DEM clock. In that case the low-pass filter should be designed accordingly, that is 200 kHz / 50 Hz = 4000. 100 nF x 4000 = 400 uF, hence the 470 uF recommendation (Nichicon UKL 470 uF/25V).
It is true that in actual music the behavior of the system around analog zero (silence and very low amplitude signal, room reflections) is more important than at high amplitudes where masking occurs. Analog zero is state change of bit #15 from 0 to 1. At musical silence there is still some dither in the recording, and bit #15 is continuously between 0 and 1. Actually the PCM code changes between 0x7FFF and 0x8000. This is the rationale behind why bit error at bit #15 (MSB) is more critical. If you like to filter it better than the other bits (having the same ripple, causing the same bit errors, but at PCM code that is at different analog signal levels), go for it.
A filter is a filter. Philips designed the DEM oscillator to operate at about 200 kHz, with filter (they call it decoupling) capacitors 100 nF. All capacitors are the same.
I analyzed in the other thread that the source impedance of the DEM pins is in the order of 10 Mohm. Also the ripple amplitude is about the same at all pins. The ripple frequency is fDEM/2 or fDEM/4 (source: Thor & John, IIRC). Consequently there is no reason to use higher capacitor at the MSB pins, because the goal is to filter the ripple amplitude below < 1/2 LSB, regardless of which bit current is being filtered. Note the DEM system works only at bits # 9 to #15, the upper 7 bits. The unfiltered ripple would cause bit errors (and artefacts in the audio band), and our goal is to push this below 1/2 LSB. There is also a static bit current error at the upper 7 bits, I deal with this in the thread referred before. At the factory the chips were being selected for static bit errors, resulting in R1, no mark, S1 grades.
If you use NOS mode and the externally forced DEM frequency is equal to WS that is 44.1 kHz, you can scale up the capacitors to 4x value, that is 4 x 100 nF. 470 nF looks a reasonable choice, for all 7 DEM filter pins. Nothing is against using higher value capacitors if you like, but you won't gain anything.
Back in this thread, John @ecdesigns experimented with 50 Hz DEM clock. In that case the low-pass filter should be designed accordingly, that is 200 kHz / 50 Hz = 4000. 100 nF x 4000 = 400 uF, hence the 470 uF recommendation (Nichicon UKL 470 uF/25V).
It is true that in actual music the behavior of the system around analog zero (silence and very low amplitude signal, room reflections) is more important than at high amplitudes where masking occurs. Analog zero is state change of bit #15 from 0 to 1. At musical silence there is still some dither in the recording, and bit #15 is continuously between 0 and 1. Actually the PCM code changes between 0x7FFF and 0x8000. This is the rationale behind why bit error at bit #15 (MSB) is more critical. If you like to filter it better than the other bits (having the same ripple, causing the same bit errors, but at PCM code that is at different analog signal levels), go for it.
Also note that at analog zero (=PCM half scale) all bits change, there is 0x7fff to 0x8000 state change. Error contribution of all bits is the same at analog zero and around it.
I'd probably suggest to add a small veroboard with a pair of 74F74 and divide BCK instead to get a higher frequency.
Capacitor size looks ok for 44.1k DEM, however I suspect getting some SMD 10nF/C0G cap's on the DEM pins to -15V can help reducing the fast edge disturbances on AOL/AOR. Of course a lot depends what analogue stage is used.
With (say) Resistor + Transformer and Tube, it's not a big concern. For Op-Amp's or solid state it is.
Thor
I'm using 3xAD844 in parallel + AD817 per channel
I was also interested in knowing the sonic effects of i2s vs PCM (sim mode) and whether sim mode is fully compatible with the DEM Grundig characteristics described above (capacitor values mentioned above) + NOS. I'm asking this because it's difficult for me to try.
Thank you all for the answers you have already given me above.
Antonio
If not using around x 4 before the pin 1, that feeds the Fdem then it is not the Grunding mode. You feed it at 44.1 k. In spite of 176,4 k.
What becomes the pin 1 behavior when in latch mode (sim mode) vs Lrck ?
Seems easier you divide from the bck pin, look at my layout. The line start from the bck towards the flip flop with a further ufl pad if wanting to do it from a masterclock (but 2 flip flop surely not enough).
What becomes the pin 1 behavior when in latch mode (sim mode) vs Lrck ?
Seems easier you divide from the bck pin, look at my layout. The line start from the bck towards the flip flop with a further ufl pad if wanting to do it from a masterclock (but 2 flip flop surely not enough).
Last edited:
Hi,
Why synch DEM to 44.1 kHz, when it invites problems which are easily avoided (44.1 kHz/4) ?
Others have mentioned the same recently wrt Grundig, the CD9009 ran 4x OS. So the direct WS synch had fDEM at 176.4 kHz.
With WS synch at 1:1 (fDEM= WS), is there available a x4 multiplier?
Any idea as to an absolute preference between dividing BCK vs MCK for DEM synch (all else being equal).
Is there any concern about the load 1541A pin 16/17 presents to either WS, BCK or MCK - if so, which is best able to 'deal with it', also are there any mitigations available which may otherwise re-order any preference ?
I wonder if there is any real world (ears) advantage to using anything more than 100nF at all decoupling positions - sounds okay to me like that (100nF) for 44.1 to 354k WS and fDEM~176.4k (470pF tied via 6k8 to -15V at both ends, hence interest in synch) - happy to try different 14x caps, but its a PITA 🙂
Cheers
Last edited:
I believe in the beginning of T. comments here, he explained well his point about lower DEM sync frequency choice. below 100 hz (iirc, not sure of this) or above 80KHz to stay outside 20 hz to 20 k hz glitches of the ears band. AKA : filtering DEM x 14 can not handle it all but is a part of the mechanism to handle that half scale bit precision the TDA1541A was able to back then (quite a performance).
As Philips and third party measurements showed clearly what Philips said already, the confortable zone is sligth before x4 44.1 K hz for the DEM clock speed to avois the best the glitches in the audio band : understand it the 14x DEM decoupling CAN NOT make 100% efficient the 7V decoupling of those 14 pins.
Some liked more the 50 Hz while being in a more noisy zone and a not optimal decoupling because the capacitance variation of hundreds of uF lythics capacitance.
With a verroboard or a pcb like I showed above and everyone can copy you can try both for real and choose what you prefer.
I think we know now all the opinions here. For myself it is clear the FDEM works best whatever the decoupling when being 4x higher than the material sampling rate. Most of the time if not high res sources or radios : 44.1 K Hz 95% of everyone old library of music materials.
I wonder why we should discuss this to death now. Iczar tastes are interresting and he had the wisdom to make it in another thread, we should drop those Fdemdiscussion there in topic to his tests as it here just discussions that slowish a little new Thorsten's paradigm (well if the listening tests confirm that, ... what we don't know for the moment as noone tried first, the discuss is yet on the dev side : PS and I/V now for the most and PCBs pending.
It is clear to me Tonimxp should use DEM x4 from the bck in spite of the pin 1 if using latch enable needed for the SIM mode.
Grunding mode was the SA70xx filter frequency (161,x Khz something frequency so almost x4 44.1 k hz which is the Philips confort zone given by the 470 pF free running DEM method aka 175.4 K hz DEM frequency something due to the variety of fabed TDA and caps variation : few picos the ears should not notice but than the scopes notice as more glitch when not locked. Again the 14x DEM frequency are not 100% efficient, sowe discuss icing on the cacke).
Btw Thorsten gave a 14x DEM decoupling road that seems efficient to filter most of the situations based on the ripples measured in real life.
At least my basic understanding...
For myself I should cope above or equal to 80 Khz he FDEM by choice but if not made before will show a pcb that permits all from Zoran idea (while it is a trade off as it is betteer to design with no vias through the layers, so a stacked pcb or trace are a little trade off from EMI and noise theory, certainlyy a small trade off, but I let the pro of the real scope not eyes but ears checked to input about this, I see a lot of critics on me but no better since now but people that drops measurements with biased post listening from that mixing what is science or not, so clearly to me no educated people. YMMV.
I am confident with Thorsten and Zoran only (John not posting anymore) on this new design because their long real world experience and knowledge about the core design arounf the TDA 1541A and I/V. Further good I/V can be more a question of personal tastes (tube/BJT/OP Amps, 100% passive) and self opinion for a part (and self ego bias for some in relation with no contextualised and precise enough measurements or simply decoralated of the sounding experience for the graphs : too much simple THD)
Because of dialectic (dialogic) : It is always good though to post questions, it can make the whole thing to progress and let think the knowledgeable people popping ups new ideas by corelation brain mecanisms.
IMHO, YMMV
As Philips and third party measurements showed clearly what Philips said already, the confortable zone is sligth before x4 44.1 K hz for the DEM clock speed to avois the best the glitches in the audio band : understand it the 14x DEM decoupling CAN NOT make 100% efficient the 7V decoupling of those 14 pins.
Some liked more the 50 Hz while being in a more noisy zone and a not optimal decoupling because the capacitance variation of hundreds of uF lythics capacitance.
With a verroboard or a pcb like I showed above and everyone can copy you can try both for real and choose what you prefer.
I think we know now all the opinions here. For myself it is clear the FDEM works best whatever the decoupling when being 4x higher than the material sampling rate. Most of the time if not high res sources or radios : 44.1 K Hz 95% of everyone old library of music materials.
I wonder why we should discuss this to death now. Iczar tastes are interresting and he had the wisdom to make it in another thread, we should drop those Fdemdiscussion there in topic to his tests as it here just discussions that slowish a little new Thorsten's paradigm (well if the listening tests confirm that, ... what we don't know for the moment as noone tried first, the discuss is yet on the dev side : PS and I/V now for the most and PCBs pending.
It is clear to me Tonimxp should use DEM x4 from the bck in spite of the pin 1 if using latch enable needed for the SIM mode.
Grunding mode was the SA70xx filter frequency (161,x Khz something frequency so almost x4 44.1 k hz which is the Philips confort zone given by the 470 pF free running DEM method aka 175.4 K hz DEM frequency something due to the variety of fabed TDA and caps variation : few picos the ears should not notice but than the scopes notice as more glitch when not locked. Again the 14x DEM frequency are not 100% efficient, sowe discuss icing on the cacke).
Btw Thorsten gave a 14x DEM decoupling road that seems efficient to filter most of the situations based on the ripples measured in real life.
At least my basic understanding...
For myself I should cope above or equal to 80 Khz he FDEM by choice but if not made before will show a pcb that permits all from Zoran idea (while it is a trade off as it is betteer to design with no vias through the layers, so a stacked pcb or trace are a little trade off from EMI and noise theory, certainlyy a small trade off, but I let the pro of the real scope not eyes but ears checked to input about this, I see a lot of critics on me but no better since now but people that drops measurements with biased post listening from that mixing what is science or not, so clearly to me no educated people. YMMV.
I am confident with Thorsten and Zoran only (John not posting anymore) on this new design because their long real world experience and knowledge about the core design arounf the TDA 1541A and I/V. Further good I/V can be more a question of personal tastes (tube/BJT/OP Amps, 100% passive) and self opinion for a part (and self ego bias for some in relation with no contextualised and precise enough measurements or simply decoralated of the sounding experience for the graphs : too much simple THD)
Because of dialectic (dialogic) : It is always good though to post questions, it can make the whole thing to progress and let think the knowledgeable people popping ups new ideas by corelation brain mecanisms.
IMHO, YMMV
Last edited:
does it affect the half scale preecision the TDA1541A is able to on some bits ? Can we hear it for real ? Measurements are first for the eyes but can bias the ears, we should make it ears check by thirs party (well educated friend with music, I used to proceed like that but it costed me a lot in wine, lol !)
cheers, continue the good work, appreciated 🙂
Not adressed to me, but my enthuusiast point here. CFA AD844 worked fine but is clearly inferior and noiser to an op861 used the transimpedance way, showed and measured Pedja Rogic. So it's possible you may like too the 1k20 R input of WalterJung AD811 Grunf refamed recently in the ad1862 thread, he made better to him than all his tube and hybrid tube I/V with ECC88 and more he experienced with for a lot of years he said. @Vunce likes it more than its own cloned Pedja's 861 double stage (I would have made it different, i.e. only one op861 as I/V and another op amp for the buffer, why not AD811 indeed here as a buffer in lieu of I/V, knowing it well in my everyday TDA1541A DAC with femto clock fifoed front end, but each eachother MMV...)
And why AD817 : filter or buffer ? Did you consider op1641/42 or op828 for instance (there are plenty more to discuss at AD catalog too) : you could be surprised. And if some mA is really needed according your upper load and cables length, the 1655/56 is certainly here a great choice too. I do use AD8597/99 as a buffer in my two stage AD1862 ( I proposed a two layers gerbers free in the thread based on front end from @miro1360 2 layers board with the JLSOUNDS board board (able of tda sim mode as well) with a smd stage for I/V and one for buffer.
I found the AD8597/99 to be alsmost as good as the AD797 in the non coloration buffer use... totally transparent, almost as all the as transparent opamps in that league: it will sound like the I/V first stage is giving plus the decoupling lythics strategy you use for the buffer (choice of model).
cheers, just my two cents.
@diyiggy : the PCM half scale, I used this term in the following meaning:
The "raw" PCM code on 16 bits is binary 0000 0000 0000 0000 to 1111 1111 1111 1111. In hexadecimal we can write 0x0000 to 0xFFFF. The TDA input format is 2s complement in I2S mode, don't mix it here.
The original analog signal let's say -1V to +1V (that is 2Vpp, about 0.71Vrms) is converted to PCM by the ADC during recording, somewhere in the early phase. Now, the -1V will be converted to PCM 0x0000 and the positive peak +1V will be 0xFFFF. The value in the middle, i.e. the analog silence or 0V will be 0x7FFF or 0x8000 (don't forget there is always some noise, and the PCM code goes up and down a couple of bits).
The half bit or 1/2 LSB precision meaning is something else. The LSB is the Least Significant Bit, in PCM representation it is 0000 0000 0000 0001 or in hex 0x0001. This is the smallest change what the code is able to resolve. It has an analog equivalent, that is for the TDA1541 it is 61 nA. On 1.5 kohm I/V resistor (data sheet opamp converter) it is 90 uV.
All bits individually should have < 1/2 LSB current precision, because the final analog output will be made of the combination af all bit current (or the lack thereof). If for example the MSB that has -2 mA contribution to the analog output, deviates by + or - 61 nA, then our actual LSB current that is 61 nA will be meaningless, and our resolution will be only 15 bits or less. If these errors of individual bit add up, or even worse some bit errors are positive, others negative, then linearity is lost. This error is DNL (Differential Non Linearity). If we consider the cumulative DNL of all 16 bits we get INL (Integral Non Linearity) that is directly related to THD what we can measure.
The "raw" PCM code on 16 bits is binary 0000 0000 0000 0000 to 1111 1111 1111 1111. In hexadecimal we can write 0x0000 to 0xFFFF. The TDA input format is 2s complement in I2S mode, don't mix it here.
The original analog signal let's say -1V to +1V (that is 2Vpp, about 0.71Vrms) is converted to PCM by the ADC during recording, somewhere in the early phase. Now, the -1V will be converted to PCM 0x0000 and the positive peak +1V will be 0xFFFF. The value in the middle, i.e. the analog silence or 0V will be 0x7FFF or 0x8000 (don't forget there is always some noise, and the PCM code goes up and down a couple of bits).
The half bit or 1/2 LSB precision meaning is something else. The LSB is the Least Significant Bit, in PCM representation it is 0000 0000 0000 0001 or in hex 0x0001. This is the smallest change what the code is able to resolve. It has an analog equivalent, that is for the TDA1541 it is 61 nA. On 1.5 kohm I/V resistor (data sheet opamp converter) it is 90 uV.
All bits individually should have < 1/2 LSB current precision, because the final analog output will be made of the combination af all bit current (or the lack thereof). If for example the MSB that has -2 mA contribution to the analog output, deviates by + or - 61 nA, then our actual LSB current that is 61 nA will be meaningless, and our resolution will be only 15 bits or less. If these errors of individual bit add up, or even worse some bit errors are positive, others negative, then linearity is lost. This error is DNL (Differential Non Linearity). If we consider the cumulative DNL of all 16 bits we get INL (Integral Non Linearity) that is directly related to THD what we can measure.
INL and DNL are explained better that I can in this article:
https://www.mvaudiolabs.com/digital/tda1541a-dynamic-element-matching-slowed-down/
Worth reading also about DEM capacitor leakage, 100 Hz DEM clock, measurements...
https://www.mvaudiolabs.com/digital/tda1541a-dynamic-element-matching-slowed-down/
Worth reading also about DEM capacitor leakage, 100 Hz DEM clock, measurements...
Thanks for the pedagogy, those calculs remind me one of my high schools perimeters when one of them was IT engineer and a headache curse (the worse of this high school in term of difficulty) about "language machine" (certainly the same words in english). Fortunally I learned aside some very different non IT fields or I couldn't survive (mentaly sane side I mean) though I'd to stay in IT for a part o my life for f my living and managing enginneers to make them walk in the same line (MOA/MOE project director) without silent gun shots in the offices was not the easier task, ahaha !
Okay, I am still an half illiterate but the question is if one is able to percieve that little gain from the THD figure among all the others coloration : understand it in engineer words when you input the signal somewhere, the output should be the same to avoid coloration (change of the signal). The question is : does this work at those low level if all the others are more changer for the input signals (your outputs traffos, ours pcb, tube and I/V stages) : That's what I try to ask and input to chase the noisest unicorns first.
Can you scale, (very interresting both academic and on the TDA1541A knowledge field) other all the rest ? It is totally off topic and I appologize, I'd like to have both your own opinion with your experiment with listening on this with your opa amp stage (op828 if my memory serves well, but it collapses unluckilly) and your sota i/V passive traffo, and if non advertised friends checked it between both with and without your DEM experiments ?
Okay, I am still an half illiterate but the question is if one is able to percieve that little gain from the THD figure among all the others coloration : understand it in engineer words when you input the signal somewhere, the output should be the same to avoid coloration (change of the signal). The question is : does this work at those low level if all the others are more changer for the input signals (your outputs traffos, ours pcb, tube and I/V stages) : That's what I try to ask and input to chase the noisest unicorns first.
Can you scale, (very interresting both academic and on the TDA1541A knowledge field) other all the rest ? It is totally off topic and I appologize, I'd like to have both your own opinion with your experiment with listening on this with your opa amp stage (op828 if my memory serves well, but it collapses unluckilly) and your sota i/V passive traffo, and if non advertised friends checked it between both with and without your DEM experiments ?
Last edited:
Short answer: I am in the middle of constructing my "ultimate NOS DAC", and I try to measure everything before listen. The NOS conversion gave big improvement, the Grundig DEM oscillator did also, WS reclockig from a Neutrino clock did it, swapping opamp ne5543 to opa828 also (all by listening myself and an enthusiast friend). Now I will try DEM sync from BCK/16, better power regulator and an I/V trafo. It just takest time.
As for the nuances in THD and bit precision: there must be some reason why the S1 and S2 grade is so high regarded. And the data sheet mentions the only difference between standard and S1 is the DNL. Perhaps there are some more differences, but I can't imagine what else. My fixa idea is THD at low level (-60 dB). This is what I can measure, but I haven't correlated it to listening yet.
As for the nuances in THD and bit precision: there must be some reason why the S1 and S2 grade is so high regarded. And the data sheet mentions the only difference between standard and S1 is the DNL. Perhaps there are some more differences, but I can't imagine what else. My fixa idea is THD at low level (-60 dB). This is what I can measure, but I haven't correlated it to listening yet.
Yup, tha's what I am understanging about your goal.
I am focusing myself more on the behavior of the thd harmonics understanding the tda has raw an even harmonic way of sounding...
I wisch i could heardyour sota traffo.... i will go bjt as it is less 3xpensive and tube for a fumier one ro look at.
Cheers.
I am focusing myself more on the behavior of the thd harmonics understanding the tda has raw an even harmonic way of sounding...
I wisch i could heardyour sota traffo.... i will go bjt as it is less 3xpensive and tube for a fumier one ro look at.
Cheers.