It's a bit more difficult.
You can calculate a maximum signal for a given frequency using the sampling rate of about 2.8Mhz. So the slew rate limit is 2.8M steps per second, but how that translates into V/µs probably depends on the player. There is no such thing as a max signal at max frequency in the DSD system.
You can calculate a maximum signal for a given frequency using the sampling rate of about 2.8Mhz. So the slew rate limit is 2.8M steps per second, but how that translates into V/µs probably depends on the player. There is no such thing as a max signal at max frequency in the DSD system.
Sorry, I'll rephrase the question.
The references below indicate that there is a native slew rate limit inherent in the DSD-SACD methodology. This is not due to the analog electronics, but rather the 1-bit DSD format and how quickly it can ramp up or down in a given time period. It seems that all we need to know is the the maximum step size in volts for a bit change, since we know what the effective DSD sample rate is. Has anyone uncovered this? Also, based on one of the articles below, it seems perhaps that the SACD algorithm may be doing some type of spline-like interpolation for the data -- this could make things a bit more complicated. Still, the slew rate limit should roughly be related to the maximum sampling rate and size of the bit step.
For background reference, please refer to the following thread:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=22446&perpage=10&pagenumber=1
and the following articles:
http://www.iar-80.com/page17.html
http://www.iar-80.com/page39.html
http://www.iar-80.com/page48.html
http://sound.westhost.com/cd-sacd-dvda.htm
I'll try to get OliverD in on the discussion, since he was the one who alerted me about this issue in the above mentioned thread...
The references below indicate that there is a native slew rate limit inherent in the DSD-SACD methodology. This is not due to the analog electronics, but rather the 1-bit DSD format and how quickly it can ramp up or down in a given time period. It seems that all we need to know is the the maximum step size in volts for a bit change, since we know what the effective DSD sample rate is. Has anyone uncovered this? Also, based on one of the articles below, it seems perhaps that the SACD algorithm may be doing some type of spline-like interpolation for the data -- this could make things a bit more complicated. Still, the slew rate limit should roughly be related to the maximum sampling rate and size of the bit step.
For background reference, please refer to the following thread:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=22446&perpage=10&pagenumber=1
and the following articles:
http://www.iar-80.com/page17.html
http://www.iar-80.com/page39.html
http://www.iar-80.com/page48.html
http://sound.westhost.com/cd-sacd-dvda.htm
I'll try to get OliverD in on the discussion, since he was the one who alerted me about this issue in the above mentioned thread...
Oops. I see OliverD has already chimed in. Have you read the iar-80 articles? They refer to a slew rate problem:
"We believe this happens because DSD-SACD cannot handle high slew rate music, and literally falls apart if asked to do so, crashing off the curvy road it can't track (see discussion in 1998 Master Guide)."
and...
"A high slew rate capability is the strength of a PCM digital system, but is the worst Achilles' heel of a 1 bit system such as DSD-SACD. If a 1 bit system cannot keep up with a fast, steeply slewing music waveform, it will grossly distort, crashing off the path of the music waveform roadway with a grotesque crashing noise of frazzled distortion -- which is precisely what we hear from the Sony-Philips DSD-SACD on cymbal sounds."
and further...
"This contrast is especially keen for a 1 bit system (e.g. DSD-SACD), since the difference in task difficulty here is primarily one of slew rate, and it is precisely slew rate which is the principal weakness of 1 bit digital systems."
Reference: http://www.iar-80.com/page19.html
So this begs the questions: Is there really an imposed slew-rate limit inherent in the SACD-DSD system due to the sampling rate/step size? ...Or is the slew rate limit refered to above not really slewing error per say, but rather a form of quantization error due to weak or error-prone averaging at high frequencies?
"We believe this happens because DSD-SACD cannot handle high slew rate music, and literally falls apart if asked to do so, crashing off the curvy road it can't track (see discussion in 1998 Master Guide)."
and...
"A high slew rate capability is the strength of a PCM digital system, but is the worst Achilles' heel of a 1 bit system such as DSD-SACD. If a 1 bit system cannot keep up with a fast, steeply slewing music waveform, it will grossly distort, crashing off the path of the music waveform roadway with a grotesque crashing noise of frazzled distortion -- which is precisely what we hear from the Sony-Philips DSD-SACD on cymbal sounds."
and further...
"This contrast is especially keen for a 1 bit system (e.g. DSD-SACD), since the difference in task difficulty here is primarily one of slew rate, and it is precisely slew rate which is the principal weakness of 1 bit digital systems."
Reference: http://www.iar-80.com/page19.html
So this begs the questions: Is there really an imposed slew-rate limit inherent in the SACD-DSD system due to the sampling rate/step size? ...Or is the slew rate limit refered to above not really slewing error per say, but rather a form of quantization error due to weak or error-prone averaging at high frequencies?
Looking at an electronic representation of music, we normally see decreasing amplitude with increasing frequency (unless emphasis is added). This helps DSD to keep track of the music waveform. It has problems following high frequency high amplitude signals, though. No doubt there are some serious issues with DSD, most of which are not discussed as widely as they should.
However, the same is true for PCM. We all assume that Shannon's theorem can be applied to PCM as long as no signal frequency higher than half the sampling frequency is present before the ADC. Which is wrong.
While I personally prefer a high resolution PCM format such as DVD-A, keep in mind that most ADCs are delta-sigma modulators and provide a PCM output only by means of a decimation filter (though they work at a higher sampling frequency than SACD).
More than 50 years of digital audio are obviously not enough to get it right
However, the same is true for PCM. We all assume that Shannon's theorem can be applied to PCM as long as no signal frequency higher than half the sampling frequency is present before the ADC. Which is wrong.
While I personally prefer a high resolution PCM format such as DVD-A, keep in mind that most ADCs are delta-sigma modulators and provide a PCM output only by means of a decimation filter (though they work at a higher sampling frequency than SACD).
More than 50 years of digital audio are obviously not enough to get it right
Unfortunately, you're right. The anti-aliasing filters used in PCM systems are designed to work at the Nyquist theoretical limit. I worked in digital sampling EEG systems where we wouldn't even consider anything as abhorent as sampling at that rate. We typically used at least 4x oversampling when digitizing the waveform. Still, this resulted in measureable phase distortion at our cut-off frequency and far below (and often aliasing due to high level, high frequency EMG signals). We should be using gentle slope cut-off filters and sampling at many times the cutoff frequency. Even if you look at 96K and 192K sampling rate filters on some equipment, they are often "brick-wall" types that cause huge problems, albeit at higher frequencies than a 20K sharp filter. And of course, reconstruction filters are another theoretical area where many seem to ignore the in-use practical problems and errors.
Nevertheless, I was and am still hoping we could dig up some information to support the iar-80 site's argument about slew distortion (as well as the author's claim of so-called 6-bit quantization) in DSD...
Nevertheless, I was and am still hoping we could dig up some information to support the iar-80 site's argument about slew distortion (as well as the author's claim of so-called 6-bit quantization) in DSD...
atalio said:
The 6-bit quantization is supposed to be the PCM data width that has the same slew rate as DSD. It is obtained from a simple calculation.
Redbook PCM uses 16-bit samples at a rate of 44,100 samples per second. It can slew 65535 steps in one sample time, 22.67us. DSD uses 1-bit samples at a rate of 2,822,400 samples per second. It can slew only 1 step in one sample time, 0.35us. A 22,050 Hz sine wave, the theoretical maximum PCM frequency, has a period of 45.34us or 128 DSD sample times. In other words, DSD can only slew 128 steps in the period of a 22.05KHz sine wave. In PCM, 128 steps can be fully encoded in 7 bits; 6 bits if you don’t count the sign bit.
DSD is banking on the fact that, in recorded music, high frequencies are very much attenuated compared to low frequencies and the inherent slew limitations will be of no consequence. The question is: What are the slew rates in recorded music? The program I wrote that analyses CD tracks for duplicate samples also collects data on the maximum slew. The highest I have found in the CDs I have examined is 96.4% of full scale, or about 63176 steps in one sample time. That’s also the calculated maximum slew of a 20KHz sine wave recorded at 0dBFS. The second highest I’ve seen is 94.4%. Slews of 60-70% are common in highly compressed recordings, such as rock and pop. Even audiophile-grade classical recordings that were recorded with lots of headroom have slews of 40-50% of full scale in one sample time. In one PCM sample time where the CD version of a recording slews 30,000 steps, the DSD version of the same recording will slew 64 steps.
Whan I would really like to test would be this :
Imagine a test setup with a multibit DAC like TDA1543 feeding a sigma-delta ADC which outputs a DSD stream.
We can then feed test signals and record the DSD stream.
This is just like recording a musical performance to DSD, except said performance will be test signals produced by the DAC.
Now, have the DAC output a test signal which would be a single sample pulse of known amplitude.
This is the absolute opposite of steady-state sinewave testing.
Then, look at the DSD output in a time window surrounding the test pulse. Suppose you look from the start of the pulse and during 2 PCM samples, this means 45 us, or 128 DSD pulses.
Now do a bit of information theory. Is DSD capable to losslessly encode PCM ? This can be easily tested. Just send pulses of all the possible amplitudes (65536 for 16 bits). Maybe send each pulse 64 times or something. Each time, record the DSD stream.
Now, if DSD can record PCM losslessly, all PCM pulses should correspond to a different DSD encoding. In this case DSD would win.
However, if several PCM pulses end up with the same DSD encoding, when replaying the DSD stream it is impossible to reconstruct the original PCM values. Thus, DSD would lose.
Personnaly I think DSD would lose big time. If it didn't, then why have all manufacturers droppped 1-bit DACs long ago and generally use multibit (generally 6 bits) sigma delta running at 6 or more MHz ?
This test can be performed in a perfectly exact way in simulation if someone can get me the equations for a DSD encoder.
Anyone ?
Imagine a test setup with a multibit DAC like TDA1543 feeding a sigma-delta ADC which outputs a DSD stream.
We can then feed test signals and record the DSD stream.
This is just like recording a musical performance to DSD, except said performance will be test signals produced by the DAC.
Now, have the DAC output a test signal which would be a single sample pulse of known amplitude.
This is the absolute opposite of steady-state sinewave testing.
Then, look at the DSD output in a time window surrounding the test pulse. Suppose you look from the start of the pulse and during 2 PCM samples, this means 45 us, or 128 DSD pulses.
Now do a bit of information theory. Is DSD capable to losslessly encode PCM ? This can be easily tested. Just send pulses of all the possible amplitudes (65536 for 16 bits). Maybe send each pulse 64 times or something. Each time, record the DSD stream.
Now, if DSD can record PCM losslessly, all PCM pulses should correspond to a different DSD encoding. In this case DSD would win.
However, if several PCM pulses end up with the same DSD encoding, when replaying the DSD stream it is impossible to reconstruct the original PCM values. Thus, DSD would lose.
Personnaly I think DSD would lose big time. If it didn't, then why have all manufacturers droppped 1-bit DACs long ago and generally use multibit (generally 6 bits) sigma delta running at 6 or more MHz ?
This test can be performed in a perfectly exact way in simulation if someone can get me the equations for a DSD encoder.
Anyone ?
DSD can slew from one end of its scale to the other in 1 sample, as can all 1 bit systems.
The stream simple goes 000000001111111111 (or the other way) and you get a step change.
The limiting factor is the analogue filtering that takes out all the HF from this signal and just passes the audio freq component.
The stream simple goes 000000001111111111 (or the other way) and you get a step change.
The limiting factor is the analogue filtering that takes out all the HF from this signal and just passes the audio freq component.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.