Hello People!
Now it's getting interesting! I just want to clarify a few points. About SACD (and high-speed delta-sigma DACs) :
if the signal is silent, the switching will be of the form:
0101010101010101010101010101....
with a 1.4Mhz square wave. This has nothing to do with what was recorded, but is simply part of the format. When there is a signal, it may appear more like:
0101101110111101110101010010100010000100000010000000...
which would hopefully contain audible frequencies, but really needs a low-pass filter to remove the ultrasound. The ultrasound I'm talking about are the sub-harmonics of 2.8Mhz, like 1.4M, 0.94M, 0.7M, 0.56M, 470k, 403k, 353k, 314k, 282k, 257k, 235k, 217k, 202k, 188k (2.8244M / 15)...... All of these (and more) have nothing to do with the recorded sound but are again just artifacts of the format. Due to the changing nature of the switching, many new combinations of frequencies are produced, and they all have positive-harmonics as well.
There several difficulties associated with the switching format:
1)Getting a good dynamic range when for example there are only a few hundred bits to work with per cycle of a sine wave. This is nowhere near the theoretical maximum of 65k or 16M amplitude variations with PCM. If we try to average the signal over a longer period of time, we introduce digital noise of a wide bandwidth.
2)The proclaimed extra bandwidth beyond 20kHz gets drowned out by digital noise. Filtering is required to protect amplifiers and speakers from these ultrasonic signals. In the context of the above, using filters with gentle slopes will do more harm than good.
3)The Nyquist Rate works differently with this type of format, and a frequency of eg: 30kHz might already create aliased frequencies below 20kHz.
4)Optimizing the switching so that most of the noise is well above 20kHz is not that simple and requires processing power (not sure how much though).
5)Jitter may not be such a big problem apart of some extra noise, but transistor slew rates can create harmonic distortion. The output is not just a sum of 1s and 0s, but the sloping values between states also add up. Ordinarily, the switching patterns would change so often and unpredictably that such distortion may be hard to measure, and the noise-floor would be more dominant.
All the latest high-speed 24bit DAC chips seem to work this way for PCM signals too. With PCM there are other difficulties:
1)Getting a semi-reasonable 20kHz tone onto a 44.1k sample/s recording requires heavy(?) digital processing to down-convert an oversampled signal from an ADC.
2)Getting a semi-reasonable delta-sigma/DSD output signal is also far from painless. Simple interpolation of the "8 times oversampling" variety is nowhere near enough, and substantial mathematical calculations are required to extract a nice 20kHz sine wave from a recording with a 22.05kHz Nyquist Rate. Try drawing a narrow sine wave on graphing paper, sample it, then try to recover the signal without reference to the original, and you'll see what I mean.
3)At a variety of frequencies, but especially high frequencies, even if the samples are extremely accurate (eg: 2^16 or 2^24 bits), that accuracy is next-to-worthless if there is even a slight amount of jitter. Not much can be done about if the inaccuracies originated in the recording equipment.
I doubt that the subjective benefits of SACD have anything to do with high bandwidth, but everything to do with the significant optimizations over PCM. Nevertheless, with the right design, I don't see anything that SACD can do that PCM can't do just as well.
Oversampling ADCs are used so that aliasing isn't a problem, but the problem gets worse again if the filtering uses a higher cutoff frequency than necessary. It's debatable if changes in phase angles due to antialiasing filtering are at all audible, if they are, then why??? Even with a 192k sample/s ADC, deliberately recording audio beyond 20kHz reduces the benefit of the high sampling rate and means that a steeper filter is needed again. Once the sound is sampled, it can be digitally filtered and down-converted to a more manageable sample rate like 44.1k without losses. Digital processing is required to recover a lot of that data, and I think that cd players mostly do a mediocre job in that respect. Simply converting the samples to analogue voltages without any processing or up-conversion is the worst possibility.
Nonlinearities in the system can include things like the dielectrics in all the capacitors and cables, semiconductors, and non-conductive materials used, as well as from magnetic storage of audio as flux in a transformer or inductor. There are other sources of non-linearity as well, but I think these are the most important ones. IM distortion (beating frequencies that turn into real tones) can result from these non-linearities, and one way to reduce it is to limit the audio bandwidth.
Phew!
CM
Now it's getting interesting! I just want to clarify a few points. About SACD (and high-speed delta-sigma DACs) :
if the signal is silent, the switching will be of the form:
0101010101010101010101010101....
with a 1.4Mhz square wave. This has nothing to do with what was recorded, but is simply part of the format. When there is a signal, it may appear more like:
0101101110111101110101010010100010000100000010000000...
which would hopefully contain audible frequencies, but really needs a low-pass filter to remove the ultrasound. The ultrasound I'm talking about are the sub-harmonics of 2.8Mhz, like 1.4M, 0.94M, 0.7M, 0.56M, 470k, 403k, 353k, 314k, 282k, 257k, 235k, 217k, 202k, 188k (2.8244M / 15)...... All of these (and more) have nothing to do with the recorded sound but are again just artifacts of the format. Due to the changing nature of the switching, many new combinations of frequencies are produced, and they all have positive-harmonics as well.
There several difficulties associated with the switching format:
1)Getting a good dynamic range when for example there are only a few hundred bits to work with per cycle of a sine wave. This is nowhere near the theoretical maximum of 65k or 16M amplitude variations with PCM. If we try to average the signal over a longer period of time, we introduce digital noise of a wide bandwidth.
2)The proclaimed extra bandwidth beyond 20kHz gets drowned out by digital noise. Filtering is required to protect amplifiers and speakers from these ultrasonic signals. In the context of the above, using filters with gentle slopes will do more harm than good.
3)The Nyquist Rate works differently with this type of format, and a frequency of eg: 30kHz might already create aliased frequencies below 20kHz.
4)Optimizing the switching so that most of the noise is well above 20kHz is not that simple and requires processing power (not sure how much though).
5)Jitter may not be such a big problem apart of some extra noise, but transistor slew rates can create harmonic distortion. The output is not just a sum of 1s and 0s, but the sloping values between states also add up. Ordinarily, the switching patterns would change so often and unpredictably that such distortion may be hard to measure, and the noise-floor would be more dominant.
All the latest high-speed 24bit DAC chips seem to work this way for PCM signals too. With PCM there are other difficulties:
1)Getting a semi-reasonable 20kHz tone onto a 44.1k sample/s recording requires heavy(?) digital processing to down-convert an oversampled signal from an ADC.
2)Getting a semi-reasonable delta-sigma/DSD output signal is also far from painless. Simple interpolation of the "8 times oversampling" variety is nowhere near enough, and substantial mathematical calculations are required to extract a nice 20kHz sine wave from a recording with a 22.05kHz Nyquist Rate. Try drawing a narrow sine wave on graphing paper, sample it, then try to recover the signal without reference to the original, and you'll see what I mean.
3)At a variety of frequencies, but especially high frequencies, even if the samples are extremely accurate (eg: 2^16 or 2^24 bits), that accuracy is next-to-worthless if there is even a slight amount of jitter. Not much can be done about if the inaccuracies originated in the recording equipment.
I doubt that the subjective benefits of SACD have anything to do with high bandwidth, but everything to do with the significant optimizations over PCM. Nevertheless, with the right design, I don't see anything that SACD can do that PCM can't do just as well.
Oversampling ADCs are used so that aliasing isn't a problem, but the problem gets worse again if the filtering uses a higher cutoff frequency than necessary. It's debatable if changes in phase angles due to antialiasing filtering are at all audible, if they are, then why??? Even with a 192k sample/s ADC, deliberately recording audio beyond 20kHz reduces the benefit of the high sampling rate and means that a steeper filter is needed again. Once the sound is sampled, it can be digitally filtered and down-converted to a more manageable sample rate like 44.1k without losses. Digital processing is required to recover a lot of that data, and I think that cd players mostly do a mediocre job in that respect. Simply converting the samples to analogue voltages without any processing or up-conversion is the worst possibility.
Nonlinearities in the system can include things like the dielectrics in all the capacitors and cables, semiconductors, and non-conductive materials used, as well as from magnetic storage of audio as flux in a transformer or inductor. There are other sources of non-linearity as well, but I think these are the most important ones. IM distortion (beating frequencies that turn into real tones) can result from these non-linearities, and one way to reduce it is to limit the audio bandwidth.
Phew!
CM
fdegrove said:Hi,
I'm afraid audio will always be very much a single person experience if you want to enjoy it to the full, regardless of tweeter type.
Put another way, having a second or third person listening in bothers me in the same way regardless of whether we're listening to a system with dome tweeters or anything else.
Having a tweeter piercing my eardrums because of beaming in the HF range bothers me even more though...
But I think Ceramicman touched an interesting point regarding in-room interaction, I often found speakers with not too wide dispersion in all axis to have more accurate pinpoint imaging.
Cheers,
Hi Frank,
To me listening to music is like having sex. I get more enjoyment out of it when I have someone to share it with. So, the times that I listen by myself are usually not by choice. Although, the phrase "different strokes for differents folks" certainly does seem to apply in this case. LOL.. just messing with you.
Later, Nick
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.