[Triode User]

Hi Rob,

Does a metal braid around the BNC cable have a similar effect on RF as the Ferrites?
Some users on ComputerAudiophile report good RF filtering results on DC/USB cables and I am guessing the same would apply to a BNC cable delivering digital data.
There is a so called "JSSG 360" cable modification; after John Swenson, which has a dual layer metal braid with a dielectric sandwiched in between but connected at both ends to create an electrical loop. The shield is not connected to the connectors at either end but "floating".

It's difficult to see how that "JSSG 360" shield could act in the same way as ferrites or as effectively as ferrites themselves. I think whole areas of electronic design and manufacture would have sussed that out if it really worked. I am slight suspicious of the whole "JSSG 360" thing anyway as the test results I have seen say it is not significantly better at shielding for RF compared to just plain braid or even braid earthed at just at one end. Beware the internet myths.

 

[Rob Watts]

Yes extra shielding will be no help whatsoever as it's nett ground currents we are trying to eliminate. This can only be done by improving the ground isolation - and that's exactly what the ferrites do, in that the coupling impedance for the ground currents is increased. If the BNC galvanic isolation was perfect, then ferrites would have no benefit... And more galvanic isolation is also increasing the ground impedance at RF (the biggest problem is around 2GHz). So ferrites are doing exactly the same as galvanic isolation, in that the coupling impedance from the two grounds (blu2 ground, Dave ground) is increased.

 

[csglinux]

Hi Rob,

There's a philosophical point I've been struggling to understand about the DAC process in general. I understand the point of the infinite response sinc function is to perfectly recreate the original bandwidth-limited analogue signal. This seems very elegant and unequivocally the only correct approach. But what if the original signal wasn't bandwidth limited? What if, hypothetically, it were the perfect truncation of a Fourier representation that stopped at, say 20 kHz (40 kHz sampling rate), of something like a step function or dirac delta function. After all, musicians these days are free to put whatever junk or garbage they like in their recordings. A good example of which would be Coldplay In this situation, is there a single unequivocally correct approach, or is any type of filtering here subject to art, interpretation, personal preferences, etc.? I can see there might be an argument for at least one of two scenarios: 1) recreate the signal perfectly up to Nyquist and live with any Gibb's phenomena, or 2) filter out any high-frequency energy in order to eliminate ringing, but without dispersion errors, and so preserve timing artifacts, 3) Is there an option 3? Is there a "right" answer here, and if so, what is it? Or if not, do you have a preference, and what is it?

And please know in advance I already feel guilty (again) for having wasted time you could have better spent on your Davina project

 

[Rob Watts]

Hi Rob,

There's a philosophical point I've been struggling to understand about the DAC process in general. I understand the point of the infinite response sinc function is to perfectly recreate the original bandwidth-limited analogue signal. This seems very elegant and unequivocally the only correct approach. But what if the original signal wasn't bandwidth limited? What if, hypothetically, it were the perfect truncation of a Fourier representation that stopped at, say 20 kHz (40 kHz sampling rate), of something like a step function or dirac delta function. After all, musicians these days are free to put whatever junk or garbage they like in their recordings. A good example of which would be Coldplay In this situation, is there a single unequivocally correct approach, or is any type of filtering here subject to art, interpretation, personal preferences, etc.? I can see there might be an argument for at least one of two scenarios: 1) recreate the signal perfectly up to Nyquist and live with any Gibb's phenomena, or 2) filter out any high-frequency energy in order to eliminate ringing, but without dispersion errors, and so preserve timing artifacts, 3) Is there an option 3? Is there a "right" answer here, and if so, what is it? Or if not, do you have a preference, and what is it?

And please know in advance I already feel guilty (again) for having wasted time you could have better spent on your Davina project

Actually this is an extremely good question, and it's a subject that I have struggled with for years, particularly with the early days of the WTA development. Going back to the 1980's, my thinking was we needed a sinc function interpolation filter, so that transient timing is recovered more accurately. And we needed 1M taps (absolutely impossible) to recover the intermediate waveform to guaranteed 16 bit accuracy.

But then the early ADC's used pretty decent analogue filters to prevent aliasing, so we did get a half decent bandwidth limited signal. But as time has gone by, people have forgotten (or more accurately chosen to disregard) basic sampling theory - and so we have the technical disasters of minimum phase, NOS, MQA on the DAC side, all creating huge image (or reconstruction) aliasing, and on the ADC side half band decimation filters creating huge amounts of decimation aliasing. And aliasing really screws up the sound quality, as transient timing is distorted by aliasing - and it doesn't matter whether it's decimation aliasing (ADC) or image aliasing (DAC side).

When the WTA algorithm was put together, the first version was limited tap length, and actually I was more concerned about the image aliasing issue - and it was focussing on that that gave the huge benefits of the WTA. But as FPGA processing power grew, and I had more taps, I found out that focussing on image aliasing was incorrect; you could have two filters, that had identical image rejection, but very very different sound quality. And I found that the closer it was to a sinc function, the better the sound quality was. But here is the rub - if you are worried about ADC aliasing, then one would expect a filter that say removed everything above 20k (including the ADC aliasing) would sound better than a sinc function, which would have more OP at 22 kHz for redbook. But no; careful listening tests revealed the sinc function type response sounded much better, even when using modern recordings that have large amounts of ADC aliasing because they use half band filters.

In fact, what actually happens is there is a balancing act between sinc function ideal, and the image aliasing issue, and that's why I spent so long fine tuning the WTA algorithm to obtain the optimum SQ. But the beauty of the WTA is that most of the filter coefficients is sinc function identical; and as the tap length increases, it converges onto the ideal sinc function. Fortunately, I have always found the SQ of the WTA is the same for all recordings - for older proper aliasing filters and modern half band types.

This whole issue is pertinent to me at the moment, as I have completed testing the Davina project decimation filters, and these have no aliasing at all. But they bandwidth limit to 21.6k for redbook: and I wonder if using this will change the sound quality in that longer tap lengths become less important. So some interesting listening awaits - but my instinctive reaction is that it won't actually change the importance of tap length. But what has been interesting, is my thinking over the last ten years has actually gone full circle back to the theoretical ideal that I thought in the 1980's. But that doesn't mean that image aliasing can't be ignored, and there is a balancing act between sinc function ideal and the need to reduce image aliasing. What is curious is that sinc function ideal is for sure the only way to perfectly recover the original bandwidth limited signal, but that how important subjectively this actually is.

 

[Currawong]

Am I thinking right that the issues above are basically the result of the physical limits of what is possible with electronics?

 

[Rob Watts]

Not really; it's more of a knowledge issue. Today we can certainly do 1M tap WTA, and with more devices even more would be possible. What we know for certain is a sinc function will perfectly reproduce the original bandwidth limited analogue signal, without any change whatsoever, so it would be as if it wasn't sampled. And the M scaler will reconstruct guaranteed to 16 bit accuracy the original bandwidth limited analogue signal; the problem of course is that modern ADC's are not effective at bandwidth limiting, being a paltry -6dB at FS/2. So we will get aliasing, and no filtering can eliminate that damage. Fortunately, the subjective evidence so far is that all recordings benefit from an M scaler - those modern half band recordings, and those older recordings from the days of proper analogue bandwidth limiting with minimal aliasing.

It's the primary reason I am working on the Davina ADC, as I can ensure that decimation occurs without aliasing (it's actually never been done before), and it will be interesting to find out a number of things - whether the act of correct bandwidth limiting causes SQ problems (the filter has a non decimation mode, where it filters but does not decimate - this made it 16 times bigger than usual and is almost as complex as an M scaler) and whether a correctly non aliasing decimation actually means I have to re-optimize the WTA filter. My strong suspicion is bandwidth limiting will have no audible consequences, and it won't change the WTA algorithm! But it's going to be very interesting to find out. The most exciting part of the project is to actually determine with absolute certainty how much ideal decimation and then M scaler reconstruction does actually change the SQ. And I will be releasing randomly named files so people can listen for themselves.

So to answer your question - no it's perfectly possible to bandwidth limit and decimate without any aliasing (unless -240 dB is important!). The issue is that ADC designers are not aware of how important aliasing is from a SQ POV. And the M scaler is the same as an ideal sinc function to better than 16 bits; so complete 16 bit recovery of the original analogue signal is for sure possible.

 

[musickid]

Hi Rob does that mean davina will prove without any doubt whatsoever whether or not the mscaler is in fact "producing" what is in effect a sound which equates to the original analogue signal as if it wasn't sampled irrespective of any bandwidth limitations. So it would be a true result for half band or the older bandwidth limited recordings that had minimal aliasing? Once this is established surely the resolving qualities of amps/transducers come into play.

What about a scenario whereby davina ensures decimation without aliasing but the mscaler does not produce an original analogue signal as if it wasn't sampled. Is this physically possible and would this mean more than 1000,000 taps are needed.

In laymans terms if i bought an mscaler i want to be sure i am hearing that analogue signal as if i wasn't sampled.

 

[Rob Watts]

It will identify the SQ losses from ideal 768 decimation, then M scaling back up to 768, so sure it will tell us how close we are. The M scaler will, in this instance, recover the original un-sampled signal to a better than 16 bit accuracy under all conditions. The only question in my mind is how accurate do we need it to be - in that is better than 16 bit enough...

 

[musickid]

Could the davina "results" actually lead to a redesigned mscaler if davina shows us the desired accuracy is not yet reached. Seems very hard to believe. Truly groundbreaking.

 

[dakanao]

Hi there Rob Watts, I send you a pm a week days ago with a question about the 3.5mm output of the Mojo (directly under the light balls) sounding a bit grainier to me than the output next right to it. And I was wondering, can that be a sign of something faulty in that jack (maybe due to pressing the 3.5mm jack of the headphone cable too hard in it), or is this maybe due to interference from the LED lights, or maybe something else?

The jack doesn't cut out to me when my headphones are plugged in, the jack just sounds a bit grainier and less defined in the bass, than the other one..

 

[Currawong]

It may be issues with the plugs themselves. Rob thought were were crazy to believe that an external amp sounded better than headphones direct from the Hugo 2, but when someone terminated their headphones with RCA plugs and said it sounded better, it seems to confirm the possibility of issues with headphone plugs and sockets not always being an ideal connection.

 

[dakanao]

It may be issues with the plugs themselves. Rob thought were were crazy to believe that an external amp sounded better than headphones direct from the Hugo 2, but when someone terminated their headphones with RCA plugs and said it sounded better, it seems to confirm the possibility of issues with headphone plugs and sockets not always being an ideal connection.

Yeah, I have to push the thick 3.5mm plug from my headphones with relative force to connect in the tight 3.5mm output of the Mojo. I'm wondering if that could have damaged the internal socket, and made it sound somehow grainy, but without it cutting the connection off from my headphones..

And if that is the case, how could it be repaired? I have bought it second hand, but am willing to pay the costs of the repair if they found a problem.

 

[Triode User]

Yeah, I have to push the thick 3.5mm plug from my headphones with relative force to connect in the tight 3.5mm output of the Mojo. I'm wondering if that could have damaged the internal socket, and made it sound somehow grainy, but without it cutting the connection off from my headphones..

And if that is the case, how could it be repaired? I have bought it second hand, but am willing to pay the costs of the repair if they found a problem.

I would have thought that it would be an easy job to replace the socket for anyone competent and shouldn’t cost much. It is not impossible that there is a slightly intermittent contact that is not giving a proper connection but there again it might not be that.

 

[dakanao]

I would have thought that it would be an easy job to replace the socket for anyone competent and shouldn’t cost much. It is not impossible that there is a slightly intermittent contact that is not giving a proper connection but there again it might not be that.

Yeah, that's why I'm not sure what the problem is. I would like to have it examinated..

 

[514077]

Deleted.