Ah sorry, my post wasn't in reply to you Dedessus rather in reply to regal saying to use 44k1 in and out.
Yeah I agree those steps are the result of the DAC doing its job just fine
I hope by now Roger is understanding that measuring stairsteps from an NOS DAC chip in hopes of getting a THD% we can all relate to is pointless. As ecdesign said the ears are part of the equipment, alternative one can use a soundard to measure the analog out since there is filtering with the ADC.
With rogers equipment what would be a "nice" test is to input his sinewave generator into the MK7 simple analog stage (at a level two give the ~2vRMS out), then he could give a meaningful THD performance of the analog stage alone and confirm that it is less than a hundredth of a percent as predicted and possible do a freq response sweep?
My concern is the +5V lifted ground causing isses, as I understand dc coupling a mofset output this way can give the lifted ground a highish impedance which can cause frequency response dips.
The other issue is obviously one would have to be very careul which amp or preamp this is coupled to if not using the circlotron. Really as a "stand alone" DAC this design must have to have coupling caps on the output when you think about it, and at that point you are better off not using the +5V ground reference all together.
Though the +5V "ground" could be very handy for using a 1:1 output transformer in lieu of a capacitor, provided the +5V supply has sufficient low output impedance to give good frequency response.
Last edited:
Hi regal,
Thanks for the answer, and tip on the soundcard, I've downloaded the software someone recommended yesterday (you, regal?) and will give that a try.
Things are now making sense and taking a bit more shape. Instead of measuring, I just *listened* to the board in the shop (a just ok system) and it sounded actually, pretty decent. So what I can see on the FFT - the aliasing artifacts - are not audible? Some of them are higher than the integers multiples of fs...don't see how this could be...
Will let you know on the sound card test.
Gary
I don't know that your FFT would even give good calculations on the square wave steps. The whole NOS theory was based on aliasing being not audible by the time it reaches the brain, you should read the paper by the Japanese fellow who came up with the NOS philosophy.
FFTs do work fine on the output of NOS DACs - I have made measurements of mine using a Sony recorder (at 96k) which I then feed into Audacity to examine the waveform captured. The FFT correctly shows the sin(x)/x roll off towards 20kHz (theory says that's -3.2dB which my FFT confirms). The envelope of the displayed waveform in Audacity is the thing not to trust - when you plot a frequency response (only when sampling at 96k or higher) the envelope of the waveform goes up, giving the appearance of rising frequency response at HF. This is an illusion which may well not appear if the NOS output is sampled at 44k1, depending on how good the anti-aliasing filter is in the soundcard. These days they're pretty good because the ADCs are in the main oversampled designs using digital AA filters.
So the upshot is - the FFT will give you the correct answer for amplitudes because its able to separate the aliasing from the real signal. Eyeballing the waveform won't, unless you took advantage of the AAF in your soundcard.
So the upshot is - the FFT will give you the correct answer for amplitudes because its able to separate the aliasing from the real signal. Eyeballing the waveform won't, unless you took advantage of the AAF in your soundcard.
BTW - Thanks to everyone that posted.
OK, I hope someone can make this clear for me...so my questions and assumptions aren't so silly.
The 1541 takes in 3 signals from the Teradak I2S: a clock signal, originally sampled at 44.1KHz, a bit clock (the system clock for operation) at 2.82MHz, and data frames/words. This clock (Fs clock) is recovered from the bitstream in I2S and will now act as the clock by which data words will get "latched in" for conversion.
So the 1541 performs a word conversion - let's say at 1KHz, and this word is of maximal amplitude (1V), as established by the incoming USB data signal.
This means that the reconstructed sine wave should have 441 steps in it, with the amplitude established by the word value. In our case, it's 1/2Fs, or about 32767. By the niceness of the 1541, it outputs this value through the R2R network: at least the first 10 bits, the last 6 establish by the proverbial DEM and decoupling caps.
So now we have an output current waveform that is apparently a reasonable reconstruct of the original, albeit with 441 quantization points. To bump it up, we amplify it through a nice FET or tube stage and since this is an active device, it makes more trouble by mixing in the staircase points with it's non-linear properties. Now this is where the tree isn't visible in the forest: is it these quantization points that cause the in-band noise (distortion, aliasing) ?
So for each point on the 1KHz "staircase", does this point create, add, contribute to the in-band noise? If so, why isn't it audible?
Is my rationale/explanation sound? If not, someone please clarify, as it would go a long way to understanding this non-audibility part. Someone else posted "...throw away any EE I know and let your ears do the filtering.." This concept I don't understand...I've always thought when it comes to signal impurities, if you can see it, measure it..you can probably hear it.
Gary
Hi all
This picture is about TDA1543 in NOS mode with our IVTCY (I/V with tube )
the blue line is 1 Khz generated with the 44.1 Khz juxtaposed
count the number of stairs
the red line use a notch filtre around 44.1 Khz
Hope it help
Hey, thanks. it does help. By using a 44.1KHz notch filter you eliminate the Fs without affecting the overall DAC signal. Good point, however, what filter topology and rolloff will work without impacting the audio signal in phase and sound quality/perception?
What did you use, is this an actual implementation, or did you do this just for a test?
Thanks,
Gary
John/EC Designs,
I appreciate you have taken the time to answer this thread, and I apologize as some of this is unfamiliar territory for me (NOS)
BTW - I have taken time to listen to what I have, and the sound is...pretty good so far. Up to now I have used my measurements to guide me, which is (from what i gather on this thread) not a good idea? Do we really need to throw out typical measurement methods with NOS?
It's still not totally clear - why are the impurities from the intermixed 44.1KHz sampling signal NOT audible in the 20-20 band? They "appear" to be high enough relative to Fs to be of concern on the scope's FFT. Even with a 30KHz LPF on the HP8903B audio test system, there's still 1.2% THD. If I were testing an amplifier or preamp, this kind of numbers would be nasty.
Can you please clarify?
Gary
So far your understanding is sound.. here's the tricky bit
Should be 44.1 quantization points for a 1kHz signal
Its the quantization that gives rise to the out of band artifacts which we call aliasing. In-band noise hopefully is very low - just the amplitude quantization noise - the band in question is just 20kHz wide.
In amplitude terms, the quantization noise is audible, just very low level indeed. Normally its masked by noise in the music you're playing. Don't confuse the results of the time quantization with those of amplitude quantization. Time quantization is giving very large out of band aliasing signals which are the result of the square boxes we've built the digital waveform from. At 1kHz they're fairly low level though coz we have many relatively small boxes building one cycle of the waveform. They turn up as images around the sample frequency and its multiples - so in this case beginning at 43k1 and 45k1. But if you imagine what happens with a 10kHz signal, you'll note that the box waveform looks very different from a 10kHz sine wave. Only 4 and a bit boxes make up one cycle. In this case the error signals turn up as higher amplitude sinewaves at 34k1 and 54k1 (and multiples of course, to infinity).
Try seeing if you can hear the difference between a 10kHz sine wave and a 10kHz square wave. You can certainly see that clearly on the scope, you can measure it. But can you hear it?
Should be 44.1 quantization points for a 1kHz signal
Its the quantization that gives rise to the out of band artifacts which we call aliasing. In-band noise hopefully is very low - just the amplitude quantization noise - the band in question is just 20kHz wide.
In amplitude terms, the quantization noise is audible, just very low level indeed. Normally its masked by noise in the music you're playing. Don't confuse the results of the time quantization with those of amplitude quantization. Time quantization is giving very large out of band aliasing signals which are the result of the square boxes we've built the digital waveform from. At 1kHz they're fairly low level though coz we have many relatively small boxes building one cycle of the waveform. They turn up as images around the sample frequency and its multiples - so in this case beginning at 43k1 and 45k1. But if you imagine what happens with a 10kHz signal, you'll note that the box waveform looks very different from a 10kHz sine wave. Only 4 and a bit boxes make up one cycle. In this case the error signals turn up as higher amplitude sinewaves at 34k1 and 54k1 (and multiples of course, to infinity).
Try seeing if you can hear the difference between a 10kHz sine wave and a 10kHz square wave. You can certainly see that clearly on the scope, you can measure it. But can you hear it?
Hi Roger57
For more information take a look here http://www.diyaudio.com/forums/digital-line-level/188708-super-common-gate-valve-i-v-converter-tda1543.html
A link to the blog of Chanmix in the page is very interesting
A capture of my version with the filtre of Alain93 (Audiyofan's member)
who replace the original notch filtre
If you understand the Molière's language have a look here Audiyofan.org • Afficher le sujet - Pré-ampli, mu-follower hybride de 6N16B but 79 pages
It's not just for test , it's my daily DAC
For more information take a look here http://www.diyaudio.com/forums/digital-line-level/188708-super-common-gate-valve-i-v-converter-tda1543.html
A link to the blog of Chanmix in the page is very interesting
A capture of my version with the filtre of Alain93 (Audiyofan's member)
who replace the original notch filtre
If you understand the Molière's language have a look here Audiyofan.org • Afficher le sujet - Pré-ampli, mu-follower hybride de 6N16B but 79 pages
It's not just for test , it's my daily DAC
Attachments
So far your understanding is sound.. here's the tricky bit
Should be 44.1 quantization points for a 1kHz signal
Its the quantization that gives rise to the out of band artifacts which we call aliasing. In-band noise hopefully is very low - just the amplitude quantization noise - the band in question is just 20kHz wide.
In amplitude terms, the quantization noise is audible, just very low level indeed. Normally its masked by noise in the music you're playing. Don't confuse the results of the time quantization with those of amplitude quantization. Time quantization is giving very large out of band aliasing signals which are the result of the square boxes we've built the digital waveform from. At 1kHz they're fairly low level though coz we have many relatively small boxes building one cycle of the waveform. They turn up as images around the sample frequency and its multiples - so in this case beginning at 43k1 and 45k1. But if you imagine what happens with a 10kHz signal, you'll note that the box waveform looks very different from a 10kHz sine wave. Only 4 and a bit boxes make up one cycle. In this case the error signals turn up as higher amplitude sinewaves at 34k1 and 54k1 (and multiples of course, to infinity).
Try seeing if you can hear the difference between a 10kHz sine wave and a 10kHz square wave. You can certainly see that clearly on the scope, you can measure it. But can you hear it?
I'm not sure I'd run a 10KHz square wave through my system, the HF drivers in the 803D's wouldn't like it !!
oops, I drop the odd zero now and then
I think...I can understand what you are saying. Here's the part that's still not clear. Why do the many DIY's and hobbyists bother running this output through a sound card (which eliminates all that in band distortion) and display the "fudged" FFT's? Are these not totally meaningless at that point? To me - if any measurements were to be used, it would be using a test instrument at the same signal that goes into the preamp (attenuator, whatever you have) before power amplification.
Thanks for your explanation - makes sense. Still do not understand why these high harmonics are not audible...if they are there at all...if I'm still sane after this...
Hi Roger57
For more information take a look here http://www.diyaudio.com/forums/digital-line-level/188708-super-common-gate-valve-i-v-converter-tda1543.html
A link to the blog of Chanmix in the page is very interesting
A capture of my version with the filtre of Alain93 (Audiyofan's member)
who replace the original notch filtre
If you understand the Molière's language have a look here Audiyofan.org • Afficher le sujet - Pré-ampli, mu-follower hybride de 6N16B but 79 pages
It's not just for test , it's my daily DAC
Great, thanks for the post. I just looked at the link and it's interesting, and maybe warrant further investigating. I was ultimately looking to use a tube stage output, but opted to test EC Designs FET stage first. Can you post (or send me) the spice model (.asc) for this? Would appreciate it so I don't have to enter it all in again.
Thanks!
Gary
No problem
You can find it in the zip file page 2 and 4
Decompress it and copy the file to your LTC's installation directory
Don't forget the component and symbol files
There is a file for testing with TDA1541 but not yet tested in
reality
For the version with TDA1543 , they work fine
in full tube mode for Alain93 and me
Yves07 build is own this day
in mixed mode with tube and mosfet for Chanmix and Kaikos
Hope it help
You can find it in the zip file page 2 and 4
Decompress it and copy the file to your LTC's installation directory
Don't forget the component and symbol files
There is a file for testing with TDA1541 but not yet tested in
reality
For the version with TDA1543 , they work fine
in full tube mode for Alain93 and me
Yves07 build is own this day
in mixed mode with tube and mosfet for Chanmix and Kaikos
Hope it help
Yes, very helpful.
Did you find that using this notch filter affects the sound? It's phase shift must be present in the audible band.
Gary
Yes of course , in my own system it affect in a good way the amplifier
without this filter , the sound is "hard" in high level frequency
the amplifier is disturbing by the stairs because it can reproduce it
Chanmix don't use filter
That depend on what is following the DAC I think
The phase shift is less a problem regarding the speaker or amplifier
"reproduction", but still audible for good ear I suppose
Hi oshifis,
With dynamic tracking I refer to the circuit ability to track complex input signals with highest possible accuracy.
Sine wave test signals with fixed amplitude can be viewed as "static" signals that are very easy to track.
Unfortunately, music contains large amount of sine waves of varying amplitude and frequency. The timbre of voices and instruments relies heavily on accurate tracking of these complex signals.
The circuit must be able to match input timing fluctuations with highest possible accuracy. So for example a 1ns signal fluctuation must be tracked with minimum errors. The higher the time tracking accuracy, the more transparent and natural the resulting sound will be.
The circuit must be able to match input amplitude fluctuations with highest possible accuracy. So for example a 1nV signal must be tracked with minimum errors. The higher the amplitude tracking accuracy, the more transparent and natural the resulting perceived sound quality will be.
With dynamic tracking I refer to the circuit ability to track complex input signals with highest possible accuracy.
Sine wave test signals with fixed amplitude can be viewed as "static" signals that are very easy to track.
Unfortunately, music contains large amount of sine waves of varying amplitude and frequency. The timbre of voices and instruments relies heavily on accurate tracking of these complex signals.
The circuit must be able to match input timing fluctuations with highest possible accuracy. So for example a 1ns signal fluctuation must be tracked with minimum errors. The higher the time tracking accuracy, the more transparent and natural the resulting sound will be.
The circuit must be able to match input amplitude fluctuations with highest possible accuracy. So for example a 1nV signal must be tracked with minimum errors. The higher the amplitude tracking accuracy, the more transparent and natural the resulting perceived sound quality will be.
Unfortunately??
Music is harmonics....
The timbre of musical instruments depends on the amount and distribution of the harmonics.....
Unfortunately??
Music is harmonics....
The timbre of musical instruments depends on the amount and distribution of the harmonics.....
Unfortunately??
Probably from techical view..
In-band distortion? I don't notice my own NOS DAC having a boat-load of in-band distortion. Its fairly clean, but not perfect.
Well to me all FFTs are 'fudged' to some extent because the FFT is designed to work with static signals. Plots that get posted up look static but they're averages. But that's a kind of side issue here so I'll stop right now
I don't see why an FFT would be totally meaningless, no. Sampled at 96k, an FFT clearly shows out of band aliasing products on NOS DACs.
Still do not understand why these high harmonics are not audible...if they are there at all...if I'm still sane after this...
Well they're not audible to humans, but perhaps bats and dogs are able to hear them
That's just one of the limitations of the human auditory system.
Is "dynamic tracking" a measurable parameter? If it is (in contrast to transparency and naturalness, which are the resulting subjective parameters), how do you measure 1ns and 1nV tracking error? I am very interested in measuring methods in general, and in particular in the correlation between measurable and subjective parameters.