In @Rugby’s Weekly Zoom meeting the other week I was very interested to hear an explanation by @ipeverywhere (Ryan) about my description of the audible difference between PCM and DSD.
I said that for me I found DSD to have advantages of a wider soundstage and better spatial cues, but there was a loss of leading edge dynamics.
Ryan noted that in a PCM stream the dB or volume information can go from zero to whatever in one cycle because it is a value within the multi-bit wide signal received each cycle. With a DSD stream, however, the dB or volume information is encoded over time within the frequency changes of the single bit so it will always take multiple cycles to transmit that information.
Now DSD uses higher frequencies than PCM so there are multiple factors affecting the time required for leading edges to be resolved. Lower DSD rates, like DSD 64 are more likely to suffer from a smeared leading edge than higher rates like DSD 256.
I was wondering whether this was consistent with other people’s experience of DSD ?
With HQ Player I find the Extra Compensation modulators and the ext2 filter reduce any leading edge issues at DSD 256 to the point where I don’t think I could pick it in a blind test.
First of all, to set the record straight, DSD is 1-bit PCM, so there are no fundamental differences between the two. Below analysis applies equally to both.
The maximum rate at which a signal can change depends on two factors: bandwidth and maximum amplitude. In a digital signal, bandwidth is limited by the sample rate (due to Nyquist limit). Amplitude is also limited due to the sample encoding. In floating point encoding, the max is by convention 1. In integer encoding, the max is the highest representable value with that bit depth, but that is roughly mapped to 1 when converting to floating point for the purpose of DSP. So, the only effective limiting factor is the sampling rate. From this point of view, DSD should fare better than some PCM due to the very high sampling rate. In fact, one of the advantages quoted by DSD proponents is a better impulse response (i.e. less ringing), which implies faster response time. That being said, the practical bandwidth of DSD is much lower than the Nyquist limit due to the sharp rising of quantization noise above a certain frequency (typically 30-50 kHz for DSD64). Thus, DSD64 should be roughly equivalent to PCM sampled at 60-100 kHz, regardless of bit depth. For higher rate DSD, those limits are of course higher.
If you encode PCM with 1 bit, you get 1-bit PCM. You can call that DSD or SDM if you like. I meant exactly what I said. You are however right that what we conventionally call PCM is not typically encoded with less than 8 bits outside of a DAC.
PCM is rarely used in 1bit structure, and not for any of the applications we’re talking about, music-related (some old video games maybe…). digital audio is LinearPCM and not DeltaPCM. so to say PCM is SDM is disingenuous - you can say it can operate in a Delta mode, but SDM, sigma-delta modulation, is the pure form of a delta digitization scheme. so. related, but not: ‘is the same as’ something that is almost never used and totally not ever in this scenario.
(just thought i’d stick my face up for a punch while others are at it… like nerd online fight club. you know, for kids.)
With HQPlayer, I initially did find ASDM7EC to sound better than ASDM7 (non EC).
But probably due to getting over excited.
I think it’s safe to say the ‘old’ ASDM7 is already a super high quality SDM modulator.
Plus it’s really important to consider the DAC itself - can you bypass it’s internal modulator or not. This can determine “the sound of DSD” quite a bit.
I know with your Holo you can do this. What DSD DAC do you use @ipeverywhere ?
But just to make myself feel better on the inside, when I can upgrade my Mac to a newer model that can run ASDM7EC while I do other tasks simultaneously , I probably will upgrade.
PCM - pulse code modulation - a binary number of n bits, greater than 1: the code – the first industrial PCM signal was 5 bits. the code is a value, a y-axis level recorded in the sampling. in DSD the output of a delta-sigma sampling process is a bitstream of 0s and 1s. there is no ‘code’ per sample frequency block, there is only a state: on or off, up or down. DSD is then pointedly, and incredibly apropos-ly, converted to PCM - for distribution to downstream formats, like CD and streaming. by definition you cannot convert a thing into itself.
Yes, there is. Each bit is a sample. Zero (or ‘off’, or ‘down’) corresponds to a level of -1 and a one (or ‘on’, or ‘up’) to a level of +1. To convert DSD to PCM, all you need to do is:
Convert every bit to a floating point sample by replacing every zero with -1.0 and every one with 1.0
Apply a low-pass filter in the digital domain.
To convert DSD to analog, all you need to do is:
For a zero, output -V; for a one, output +V
Apply a low-pass filter in the analog domain.
Yes, you can You can up-convert or down-convert PCM, or change bit depth.
But I know this can get very complicated so sometimes you need to simpify to keep the discussion simple also. Otherwise it can get really deep ‘in the weeds’ and only a handful of people can follow what is happening
It is not really true, because single PCM sample rarely coincides your signal peaks. This is why for example NOS DACs running at low sample rate work very badly. You need multiple PCM samples and sinc-function to reconstruct the analog signal. Especially when your signal is close to Nyquist frequency of the PCM sampling used. Your transient (step response step) can reside anywhere between two sample points. Many PCM DACs have incomplete reconstruction and thus cannot correctly reconstruct transients encoded in the PCM data.
Non-Nyquist sampling systems like DSD64 can encode frequencies well above 100 kHz. I have earlier shown 250 kHz sine waves reproduced by DSD64 DAC. Just your SNR at 250 kHz gets poor with DSD64.
Bandwidth ultimately defines your transient response. For example DSD256 is enough to provide over 24 PCM-bit worth of dynamic range for 100 kHz bandwidth. For DSD512 this is 200 kHz and for DSD1024 this is 400 kHz.
With any SDM, output quality essentially depends pretty much 50/50 on the modulator and the D/A conversion (analog reconstruction filter).
Main issue with PCM D/A converters (R2R etc) is settling time vs analog reconstruction filter. To accurately reconstruct for example 24-bit resolution, the DAC output would need to settle within ±½ LSB in fraction of sample period. This limits possible sampling rate. While having limited sampling rate means difficulty to create analog filter that would remove all images to correctly reconstruct the output signal. If there’s high order analog filter with low corner frequency, it in turn spoils phase response of the audio band. Also the true conversion linearity is usually between 15 and 20 bits. For these reasons, world has moved on with SDM converters. DSD/SDM/PWM ultimately solves this linearity and settling time issue because there are only two values to settle to.
the arguments and explanations you present are spurious and misleading. no one asked about the … well all that stuff you just spouted. it’s just noise. look up any process related to DSD - it gets converted to PCM. cause it’s PDM, cause it’s a relative state bitstream. not a coded level. so. keep talking. changing sample rate and bit depth isn’t changing from basic encoding method, as is PDM/DSD to PCM. so, that, too, is a spurious ‘point’ you make.
it’s my fault for getting into it [moderated] i was high, it was late, i was bored. i hereby bow out.
