Op Amps in CD3.5
- Thread starter Specul8tr
- Start date
martin clark
pinko bodger
They are all parts of the reconstruction filter. First two are each configured as 3-pole Sallen-Key lowpass filter, final one buffers the output, iirc. There is also a single low-pass pole built in to the output opamp in the TDA1305 dac, so the whole setup is a 7-pole filter of Bessel alignment, with an output buffer.
Julf
Facts are our friends
The primary one is the reconstruction filter - essential for recreating the original waveform.
martin clark
pinko bodger
Naim opted for a 7-pole filter in order to get the necessary image rejection as a proper reconstruction filter; and it takes a huge bite out of out-of-band noise. If you model it, you find it’s a neat bit of work; and having essentially solved the problem once, the approach got used in all Naim’s later players - albeit with slight coefficient tweaks to complement the digital filter used.
Many manufacturers use fewer poles, or even wilfully bizarre approaches on grounds of ‘sound quality’ or just plain cost cutting). Naim, I think, got this right from the outset.
Many manufacturers use fewer poles, or even wilfully bizarre approaches on grounds of ‘sound quality’ or just plain cost cutting). Naim, I think, got this right from the outset.
martin clark
pinko bodger
Can;t really comment on that, I haven't played in years.
When I did try the LME49710 (single version of the lm4562) and thought them horrible - functionally worked ok, measured ok, but turn out to be v sensitive to rf on the inputs, perhaps why it sounded staggeringly 'flat'. Not a drop-in replacement! (not least because its a bipolar input and the opamps used are FET types, with good reason...)
When I did try the LME49710 (single version of the lm4562) and thought them horrible - functionally worked ok, measured ok, but turn out to be v sensitive to rf on the inputs, perhaps why it sounded staggeringly 'flat'. Not a drop-in replacement! (not least because its a bipolar input and the opamps used are FET types, with good reason...)
Specul8tr
pfm Member
Julf, could you explain how this works?
martin clark
pinko bodger
It’s a result from Shannon’s sampling theorem - the input signal conversion to digital is based on a limited sampling bandwidth; on conversion back to analogue a reconstruction filter removes the spuriae above this same bandwidth to leave just the reconstructed input signal. An essential complement, in the mathematical sense.
Julf
Facts are our friends
What Martin said. Here is an illustration:
The red is the recorded digital signal - samples at discrete points in time. The reconstruction filter interpolates between the points to reconstruct the original waveform (gray). Forget the "staircase" pictures, they are very misleading (and actually wrong).
darrenyeats
pfm Member
Note in a typical DAC the reconstruction (anti-imaging) filter occurs in stages. The reality is more complicated, but here's the basic idea.
1. The input is oversampled to a higher frequency, e.g. 384kHz, then a digital filter is applied.
2. Then an analogue filter is applied, e.g. desired stop-band >= 192kHz.
The purpose of (1) is to allow the analogue filter (2) to kick in at a higher frequency, with a shallower slope, making (2) easier, cheaper, more benign etc.
Although it's tempting to think filtering at (1) is mathematically perfect with zero error, in fact you can see digital filter errors in FR ripples and other distortions (I can give examples). This is not necessarily audible; the intent is for the end result to better a pure analogue filter.
Newer DACs with more digital processing power have opportunity for improved measured performance via better digital filtering. Again, not necessarily audible.
1. The input is oversampled to a higher frequency, e.g. 384kHz, then a digital filter is applied.
2. Then an analogue filter is applied, e.g. desired stop-band >= 192kHz.
The purpose of (1) is to allow the analogue filter (2) to kick in at a higher frequency, with a shallower slope, making (2) easier, cheaper, more benign etc.
Although it's tempting to think filtering at (1) is mathematically perfect with zero error, in fact you can see digital filter errors in FR ripples and other distortions (I can give examples). This is not necessarily audible; the intent is for the end result to better a pure analogue filter.
Newer DACs with more digital processing power have opportunity for improved measured performance via better digital filtering. Again, not necessarily audible.
Last edited:
Specul8tr
pfm Member
The 'spuriae' presumably being jitter induced?
Julf
Facts are our friends
No. It has nothing to do with jitter. It it is the imaging at frequencies above the Nyquist limit, intrinsic to the process. It is part of the transformation of discrete dirac impulses (the samples of the audio waveform at zero-length points in time) into a continuous wave - formally it is the application of the Whittaker-Shannon interpolation formula.
Specul8tr
pfm Member
And that's why I love this forum. So much knowledge within...
So, while I can't pretend to understand the maths, I do get the general idea. Oversampled; bunch of high frequencies undesirable but necessary (or at least v. important for accuracy); digital filter within TDA1305; 2 analogue op amp low pass filters down to circa 20KHz; final op amp (unity gain, low impedance?) buffer to outside world. Simple, but effective.
So, while I can't pretend to understand the maths, I do get the general idea. Oversampled; bunch of high frequencies undesirable but necessary (or at least v. important for accuracy); digital filter within TDA1305; 2 analogue op amp low pass filters down to circa 20KHz; final op amp (unity gain, low impedance?) buffer to outside world. Simple, but effective.
Julf
Facts are our friends
To be really precise, the oversampling is not necessary (just nice), the filters would still be needed and the DAC could be just as accurate (the reason a non-oversampling DAC isn't as accurate has to do with how it is hard to make really steep analog filters, and linearity issues we probably don't want to get into).
martin clark
pinko bodger
This is exactly right: and the bit that is not immediately intuitive is simply this: the necessary 'interpolation' comes simply from 'low pass filtering' i.e. chucking-away everything in the raw DAC output that is above the input's pass band (c.21khz , for CDs.)
It is not (as reviewers often posit) some separate, semi-magical, or proprietary, even sentient 'smoothing' - just an entirely-determined and intrinsically-defined process of conversion from analogue signal, to digital representation, back to reconstructed analogue signal. That's it, and that's all.
(yet we are still left with holy wars about the shape, nature and flavour of how the very last part 'should' be done, from the ignorant 'NOS 16/44.1' heresy to 'apodising' which conflates effects of the digital oversampling filter with manufacturing desire to get good measurements in magazines & use the cheapest-possible analogue output stage, but claim a 'win' in perceptual BS terms. It's late, and I'm too tired to wade-in there...)
Specul8tr
pfm Member
‘Holy Wars’ are all well and good but all I care about is whether it sounds good or not. Well actually, that’s not quite true - I’m interested in why it sounds good too, and Martin, Julf, your comments are much appreciated and understood. I have been busy reading up on Sallen-Key, Bessel and Butterworth filters (Art of Electronics et al) and trying to relate what I’m reading to the actual layout and specific applications therein.
Why am I bothering with an age old CD player? Because it’s what I’ve got and I love it.
mega lord
Centre tapped
I love it too.
Have you changed all the opamps to opa627 yet Stephen ? Martin reckons the anologue stage is pretty well sorted so not sure what you would be wanting to do to it - or just understand what it does ?You could always lampizate it with a nice valve output stage
Only kidding.
Specul8tr
pfm Member
Don't think I haven't considered the Lampization! Trouble is, I'm not sure I dare deal with the massively high voltage involved. However, I also don't understand how and why that would improve things and would prefer to understand the existing circuit before implementing such a radical change. And there also lies the answer to your second question - I'm simply trying to understand the process the signal has to go through right from the very start of the chain to the very end. I also get the impression that op-amps are somewhat frowned upon in audio circles and if that is the case, then why would a company like Naim use them?
Another answer to your question would be that my attempts to decipher op-amp circuits is simply the next step in my attempt to understand the world of electronics as I find it thoroughly fascinating but equally bewildering.
And as for OPA627s, yes. A long time ago, which is why I was wondering whether technology has moved along far enough to consider replacing them.
Another answer to your question would be that my attempts to decipher op-amp circuits is simply the next step in my attempt to understand the world of electronics as I find it thoroughly fascinating but equally bewildering.
And as for OPA627s, yes. A long time ago, which is why I was wondering whether technology has moved along far enough to consider replacing them.
Julf
Facts are our friends
They are frowned on by audiosnobs who are still passing on old wive's tales from the 1970's. Most top engineers and designers are more than happy to use opamps these days - for a good reason. Having the whole opamp on one small piece of silicon helps deal with HF, thermal effects and interference much better than spreading it out as discrete components laid out on a circuit board.