Hi all
I'm not sure if this is the right place to ask, but I would like to either find or generate myself some test tones that can be used from CD or memory stick for challenging intermodulation distortion test tones that produce in-band residuals. So 44.1kHz 16-bit AIFF files would appear to be the most generally usable.
I am thinking about e.g. Bob Cordell's MIM tests (either the originals or shifted up to the top of the audio band as described here https://www.diyaudio.com/forums/sol...nterview-error-correction-18.html#post1047488) or Dick Small's TDFD as described in his 1986 AES article.
They would be useful for a number of things, in particular allowing any digital playback device to become a sophisticated and repeatable test signal generator, but also to allow testing of the D/A itself, then of course through the line level and power amplifier chain from there.
If I was doing it myself, I would create the tests at different levels from maximum down in 10dB steps, with and without 1 LSB triangular dither.
If these files are already available somewhere then please point me there. Otherwise I will have to read up on the AIFF format and generate some myself from first principles. I would generate the signals in 64-bit double precision and then use Stochastic Rounding (essentially an unbiased method for getting from DP to 16-bit integer that is closely related to dither see e.g. [1904.11263] Stochastic rounding and reduced-precision fixed-point arithmetic for solving neural ordinary differential equations ). Both the SR and the dither signal would be generated from a top quality PRNG such as Marsaglia's KISS64.
Any thoughts or pointers welcomed - especially if I have missed something obvious!
M
I'm not sure if this is the right place to ask, but I would like to either find or generate myself some test tones that can be used from CD or memory stick for challenging intermodulation distortion test tones that produce in-band residuals. So 44.1kHz 16-bit AIFF files would appear to be the most generally usable.
I am thinking about e.g. Bob Cordell's MIM tests (either the originals or shifted up to the top of the audio band as described here https://www.diyaudio.com/forums/sol...nterview-error-correction-18.html#post1047488) or Dick Small's TDFD as described in his 1986 AES article.
They would be useful for a number of things, in particular allowing any digital playback device to become a sophisticated and repeatable test signal generator, but also to allow testing of the D/A itself, then of course through the line level and power amplifier chain from there.
If I was doing it myself, I would create the tests at different levels from maximum down in 10dB steps, with and without 1 LSB triangular dither.
If these files are already available somewhere then please point me there. Otherwise I will have to read up on the AIFF format and generate some myself from first principles. I would generate the signals in 64-bit double precision and then use Stochastic Rounding (essentially an unbiased method for getting from DP to 16-bit integer that is closely related to dither see e.g. [1904.11263] Stochastic rounding and reduced-precision fixed-point arithmetic for solving neural ordinary differential equations ). Both the SR and the dither signal would be generated from a top quality PRNG such as Marsaglia's KISS64.
Any thoughts or pointers welcomed - especially if I have missed something obvious!
M
Last edited:
If single precision and .wav would do, you could very easily create those files with the GoldWave audio editor using its expression evaluator: add a couple of sine waves and two random numbers for dithering. Maybe it also supports AIFF, never tried that.
Thanks @MarcelvdG I'll see if that is on my platform (Mac). I am pretty fussy about dither and the quality of PRNGs and conversion from floating-point to integer so I would have to have faith that they have considered all these issues carefully.
I only suggest AIFF because I would like to be able to put them on a test CD to use within CD players as well as a more general data files.
I only suggest AIFF because I would like to be able to put them on a test CD to use within CD players as well as a more general data files.
Thanks @JohnPM. I have that on my Mac for doing loudspeaker and room measurements but I didn't realise that it had those capabilities too. I need to investigate it further but clearly it's a pretty powerful piece of kit, especially when you think that it's free to download.
If you put WAV files on a CD I assume that a normal CD player won't recognise them. Maybe load them into some software and export as AIFF?
If you put WAV files on a CD I assume that a normal CD player won't recognise them. Maybe load them into some software and export as AIFF?
Just checking my version of REW (5.19) which is a fresh download and it has the Dual Tone but not Triple Tone options.
With Windows computers, you can usually take 16 bit 44.1 kHz stereo .wav files and tell a CD-R writing program to put them on a CD-R and to make it an audio CD rather than a CD-ROM. Most regular CD-players can then read the CD-R with no problem. I never had to use AIFF to get a CD-R that plays on audio CD-players - in fact I never heard of AIFF before today.
IMO it's the Apple ecosystem, using apple proprietary formats. The burnt CD-DA medium will be identical for all OSes.
So creating the WAV files on REW, converting them to Apple AIFF (AIFF-C which I think is little-endian) in Audacity and then importing them into iTunes works and is painless. That way I have generated a bunch of test signals in a playlist and then burnt them to an audio CD which works perfectly. If anyone wants these files I can share them.
I have established that my Meridian 203 DAC has a cleaner noise floor and better harmonic structure from a variety of THD and IMD tests than the audio out of my re-capped Marantz CD94. So it would be great if it sounded better too but it really doesn't. The stereo depth is noticeably shallower on good recordings and there is a kind of 'grey veil' laying over the detail in a recording. I'm comparing with levels set very precisely and the DAC is taking either optical or electrical output from the CD94, so the same bitstream is being used. Perhaps there is something wrong with it, but it measures fine. Stereophile used to think it was a great DAC.
I'm not somebody who listens to cables and opamps because having run many ABX tests and other listening tests in carefully controlled environments whilst at Rogers and KEF and then as a consultant since (also less formally as a recording engineer), I have seen what people describe as 'night and day' differences disappear when they are made to listen double blind with levels set precisely. If one cannot reliably hear a difference then a perceived preference is irrelevant. This makes ABX tests a very powerful tool which should be used more often than they are (see e.g. Hopkins Research - ABX methods for measuring human perception of change). Unfortunately, the human motivations for profit and/or guru status are too powerful for people to forego. However, there is clearly some way to go finding measurements that correlate reliably with subtle aspects of sound quality.
I have established that my Meridian 203 DAC has a cleaner noise floor and better harmonic structure from a variety of THD and IMD tests than the audio out of my re-capped Marantz CD94. So it would be great if it sounded better too but it really doesn't. The stereo depth is noticeably shallower on good recordings and there is a kind of 'grey veil' laying over the detail in a recording. I'm comparing with levels set very precisely and the DAC is taking either optical or electrical output from the CD94, so the same bitstream is being used. Perhaps there is something wrong with it, but it measures fine. Stereophile used to think it was a great DAC.
I'm not somebody who listens to cables and opamps because having run many ABX tests and other listening tests in carefully controlled environments whilst at Rogers and KEF and then as a consultant since (also less formally as a recording engineer), I have seen what people describe as 'night and day' differences disappear when they are made to listen double blind with levels set precisely. If one cannot reliably hear a difference then a perceived preference is irrelevant. This makes ABX tests a very powerful tool which should be used more often than they are (see e.g. Hopkins Research - ABX methods for measuring human perception of change). Unfortunately, the human motivations for profit and/or guru status are too powerful for people to forego. However, there is clearly some way to go finding measurements that correlate reliably with subtle aspects of sound quality.
Have you tried playing 1 Hz sine waves of various levels and listening to the DAC output via a high-pass filter that blocks the 1 Hz?
According to Meridian 203 D/A processor Robert Harley page 1 | Stereophile.com , the Meridian 203 uses Philips SAA7321 Bitstream chips, and according to Evolution of DAC & digital filter - DutchAudioClassics.nl, those use second-order noise shaping. Noise shapers, especially those of low order, can have some funny artefacts at low signal levels. For example, idle tones that get frequency-modulated by the signal. Those can be moved out of the audio band by adding an offset, but signals that are equal but opposite to the offset will then move them back in. No idea if Philips used tricks like that, but they might have.
Another thing standard tests don't usually catch is the effect of intersample overshoots. Some CDs have the largest samples normalized to full scale, but the signal may then have to exceed full scale in between the samples. Digital interpolation filters are often not designed to handle that. (You could argue that CDs should not be recorded like that, but unfortunately some are.) The attachment contains some test files for this effect.
According to Meridian 203 D/A processor Robert Harley page 1 | Stereophile.com , the Meridian 203 uses Philips SAA7321 Bitstream chips, and according to Evolution of DAC & digital filter - DutchAudioClassics.nl, those use second-order noise shaping. Noise shapers, especially those of low order, can have some funny artefacts at low signal levels. For example, idle tones that get frequency-modulated by the signal. Those can be moved out of the audio band by adding an offset, but signals that are equal but opposite to the offset will then move them back in. No idea if Philips used tricks like that, but they might have.
Another thing standard tests don't usually catch is the effect of intersample overshoots. Some CDs have the largest samples normalized to full scale, but the signal may then have to exceed full scale in between the samples. Digital interpolation filters are often not designed to handle that. (You could argue that CDs should not be recorded like that, but unfortunately some are.) The attachment contains some test files for this effect.
You can look at a 1 Hz sine wave as a DC level that slowly gets swept up and down, so the idle tone frequency will be swept as well, and twice per second, the sine wave will be close to the opposite of any deliberately added DC offset.
Why not use an audio test software package to output the test signal from your computer soundcard? Seems way more convenient to me.
That said, Audacity can be used to create and export many wave forms. If that is not enough, Matlab or Octave will do ANY test signal you want, and you can export to AIFF (MATAA has a special tool for this, but you can do without it).
That said, Audacity can be used to create and export many wave forms. If that is not enough, Matlab or Octave will do ANY test signal you want, and you can export to AIFF (MATAA has a special tool for this, but you can do without it).
A 1 Hz test at 0 dBFS can show if the noise floor gets modulated by the signal: do you just get very soft background noise or a woosh-woosh-like sound?
A 1 Hz test at a low level (I'd try -30 dBFS, -40 dBFS and -50 dBFS) can show whether the noise floor really sounds like noise, or whether there are whistles that get frequency-modulated by the 1 Hz signal.
I'll see if I can write a more elaborate explanation later.
Thanks @MarcelvdG. I have limited experience with correlating measured and audio performance of DACs but I am keen to learn more about it. I have to decide whether to use this Meridian 203 as a basis for more tests (e.g. different output methods from the DAC - I particularly like the idea of a high quality balanced output transformer from Sowter, Jensen or Lundahl) or whether these early bitstream DAC and filter implementations are just fundamentally flawed in terms of top quality audio. The attached circuit Meridian designed looks pretty good to me though: passive 2nd order filter before the opamp to avoid slew rate issues, a 'true' differential input to cancel common mode noise and a DC servo to avoid any electrolytics in the signal path. Otherwise I will sell it to an enthusiast on eBay and put the cash into a more modern board and experiment with that. It would be nice to have USB input as well.
In case anyone is interested, I have attached some screenshots from REW using an Alesis io2 input for 1kHz tone and CCIF two tone distortion measurements at -10dB straight out of the 203 from my test disc on the CD94 transport. The measurement setup is a little crude but these look OK to me, certainly less noisy than the CD94 audio output which looks a lot noisier but sounds so much better.
In case anyone is interested, I have attached some screenshots from REW using an Alesis io2 input for 1kHz tone and CCIF two tone distortion measurements at -10dB straight out of the 203 from my test disc on the CD94 transport. The measurement setup is a little crude but these look OK to me, certainly less noisy than the CD94 audio output which looks a lot noisier but sounds so much better.
Attachments
I also have only limited experience with correlating measured and audio performance of DACs, but I have worked on sigma-delta modulators for various applications, so I know something about their weak points. Attached is my attempt at writing the more elaborate explanation that TNT asked for.
By the way, regarding section 5 of the attachment, a hobby DAC I made clearly shows noise modulation when driven with a 1 Hz, 0 dBFS sine wave, but I never heard it on music. Besides, it was well-received during a meeting of Dutch audio enthusiasts, and by two British people from this forum who made their own version of it.
By the way, regarding section 5 of the attachment, a hobby DAC I made clearly shows noise modulation when driven with a 1 Hz, 0 dBFS sine wave, but I never heard it on music. Besides, it was well-received during a meeting of Dutch audio enthusiasts, and by two British people from this forum who made their own version of it.
Very interesting. I guess the DAC in a 203 (SAA7321) being a very early delta-sigma design is single bit and therefore most prone to these modulations.
Do you have any recommendations for current designs that have learned all these lessons so that they (1) sound good and (2) measure well? ...And (3) are amenable to experimenting on a quality board design that is affordable via eBay or the like?
Do you have any recommendations for current designs that have learned all these lessons so that they (1) sound good and (2) measure well? ...And (3) are amenable to experimenting on a quality board design that is affordable via eBay or the like?
I guess so to, but I'm not entirely sure. Some early Philips designs just used a plain old single-bit sigma-delta modulator, but some others used a combination of a sigma-delta and a pulse width modulator. With such a combination, the sigma-delta modulator itself is multibit and can in principle be dithered according to dither theory to get rid of its low-level artefacts. However, dither theory was still under development at the time, so it is doubtful if they actually did that. When you don't dither a multibit sigma-delta modulator properly, you get the same issues as with a single-bit sigma-delta modulator.
I can't give you any recommendations for a board, I hope some other members will chime in.
I can't give you any recommendations for a board, I hope some other members will chime in.