Actually, this is rather a diversion, and relatively meaningless. Magnetic tape appears to capture a lot of low level information that appears to get lost in digitizing with a relatively low number of bits. Now, if low level information was digitized with 16 bit accuracy, rather than 1 or 2 bit accuracy, perhaps one major difference between analog and digital recording would be resolved, but that is not resolved at the moment.
I would like to point out that Doug Sax has access to some of the best digital recording equipment available, and he has an ear for the live sound as well. If digital did it 'right', he would give up on analog equipment, and even referring to it.
I would like to point out that Doug Sax has access to some of the best digital recording equipment available, and he has an ear for the live sound as well. If digital did it 'right', he would give up on analog equipment, and even referring to it.
John Eargle, not exactly a slouch at recording, disagreed with that. But he understood what dither was about and how it worked. 16 bits sets a limit to the noise floor, but not to resolution.
I might add, that for the past 50 years, I have exclusively owned and driven 4 cylinder autos, including 4 sports cars, and 4 passenger cars. Yet, my associate, Jack Bybee drives a 12 cylinder car. (He also owns a Blowtorch). The reasons for 12 cylinders is for power and smoothness in response. It is very much the same with the Blowtorch. For the record, the major heat generating components are a dual shunt regulator, and the output stage of the Blowtorch, both circuits made of T0-220 power mosfets. To operate these devices at low milliamp levels, adds distortion, that can be easily measured.
If I could find a complementary mosfet pair that was happy at 15 ma or so, and could be easily attached to a heatsink, I would use it.
If I could find a complementary mosfet pair that was happy at 15 ma or so, and could be easily attached to a heatsink, I would use it.
Not at all. I happened to have ordered have a kit from Erno Borbely that fits that description--shunt regulators and T0-220 MOS-FETs for the outputs. I gave up on paralleling a bunch of JFETs for a headphone amplifier and also on using batteries for power supply (both of which I currently use).
.
Magnetic tape resolution
Agreed, John. In the late 1980s and early 1990s I was doing record circuitry development in the cassette replication field, and was involved with the design of a digital high-speed master loop-bin system. At 80:1 duplication ratio, the audio band of 20Hz to 20kHz is spread over 1.6KHz to 1.6MHz (you want to talk about wide bandwidth circuitry!). Respecting the Nyquist limits required a D>A clock speed of 3.5MHz minimum. At that time, a 16-bit 3.5MHz hybrid D>A was pricy indeed! (thousands of $$) The 12 and 14-bit parts were MUCH cheaper, and, according to a competetor, sufficient for a low-fi medium like a cassette.
In response I developed a presentation delivered at the yearly ITA conference which was an examination of just how many bits of resolution were needed to optimally utilize the dynamic range of a cassette. The tests I did utilized an HP Spectrum Analyzer and single tones at various levels and frequencies in addition to my ears and a pair of Stax headphones in a quiet room.
What I found reinforced a basic fact that many engineers (myself included
) sometimes forget: Specifications like Dynamic Range, S/N, etc are single summary numbers which for many purposes overly-simplify the reality they claim to represent.
Using the Dolby™ B-Type codec, a single tone at 3KHz could be easily discerned in the noise floor at -115dB record level re: 250nWb/M. The HP also clearly showed the energy above the noise floor at 3KHz. A casual examination reveals why; at the -60dB noise level specification quoted for a cassette track at 1.875IPS, the total energy in the noise floor close enough in frequency to mask 3KHz is as much as 60dB lower. This masking range is of course referred to as a critical band, and is quite variable from individual to individual depending on the density of sensory cells in the cochlea. (this is why I do not trust the "average" masking profile upon which mp3 and other lossy data reduction codecs are modelled). BTW, a good 75% of everyone at that ITA conference could hear the -115 tone in the noise floor as well.
The result of my investigation was that (dither as a separate subject) 14 bits was definitely not sufficient, indeed even 16 bits was not enough to fully utilize the dynamic range of even an analog compact cassette. Apparently others were listening as well, we ended up making cassettes for Telarc, DMP and also the Sheffield cassettes for Doug Sax, with whom I spent some very interesting times discussing audio. I really appreciate his minimum component topology approach to recording. A simple listen shows that it works, listening to the playback of one of his analog reel-to-reel recordings on his studio system was a revelation!
Moral of the (too long) story: I have since learned to be very circumspect when appraising specifications. The fullness of the reality is in the details...
Keep the good info coming!
Howard Hoyt
Agreed, John. In the late 1980s and early 1990s I was doing record circuitry development in the cassette replication field, and was involved with the design of a digital high-speed master loop-bin system. At 80:1 duplication ratio, the audio band of 20Hz to 20kHz is spread over 1.6KHz to 1.6MHz (you want to talk about wide bandwidth circuitry!). Respecting the Nyquist limits required a D>A clock speed of 3.5MHz minimum. At that time, a 16-bit 3.5MHz hybrid D>A was pricy indeed! (thousands of $$) The 12 and 14-bit parts were MUCH cheaper, and, according to a competetor, sufficient for a low-fi medium like a cassette.
In response I developed a presentation delivered at the yearly ITA conference which was an examination of just how many bits of resolution were needed to optimally utilize the dynamic range of a cassette. The tests I did utilized an HP Spectrum Analyzer and single tones at various levels and frequencies in addition to my ears and a pair of Stax headphones in a quiet room.
What I found reinforced a basic fact that many engineers (myself included
) sometimes forget: Specifications like Dynamic Range, S/N, etc are single summary numbers which for many purposes overly-simplify the reality they claim to represent. Using the Dolby™ B-Type codec, a single tone at 3KHz could be easily discerned in the noise floor at -115dB record level re: 250nWb/M. The HP also clearly showed the energy above the noise floor at 3KHz. A casual examination reveals why; at the -60dB noise level specification quoted for a cassette track at 1.875IPS, the total energy in the noise floor close enough in frequency to mask 3KHz is as much as 60dB lower. This masking range is of course referred to as a critical band, and is quite variable from individual to individual depending on the density of sensory cells in the cochlea. (this is why I do not trust the "average" masking profile upon which mp3 and other lossy data reduction codecs are modelled). BTW, a good 75% of everyone at that ITA conference could hear the -115 tone in the noise floor as well.
The result of my investigation was that (dither as a separate subject) 14 bits was definitely not sufficient, indeed even 16 bits was not enough to fully utilize the dynamic range of even an analog compact cassette. Apparently others were listening as well, we ended up making cassettes for Telarc, DMP and also the Sheffield cassettes for Doug Sax, with whom I spent some very interesting times discussing audio. I really appreciate his minimum component topology approach to recording. A simple listen shows that it works, listening to the playback of one of his analog reel-to-reel recordings on his studio system was a revelation!
Moral of the (too long) story: I have since learned to be very circumspect when appraising specifications. The fullness of the reality is in the details...
Keep the good info coming!
Howard Hoyt
Last edited:
Interesting post, thanks. But why would you keep dither as a 'separate subject'? - of course with no dither we'd need a vastly increased bit depth, even then we'd still get quantisation distortion. But that's like saying of analog tape <with bias as a separate subject>. No reasonable engineer uses tape without bias, similarly no reasonable engineer suggests digital without dither.
Or am I missing something really vital here?
Although I think your analogy is a bit strained, I agree with the intent of your statement. At the time, there were many different approaches to dithering, involving different spectral noise shaping, each which gave different S/N readings on conventional instrumentation. The additional confounding factor was that there was not a similar dither available for both 14 and 16 bits at the time, which masked the pure difference between 14 and 16 bit. What we were attempting to measure was the difference between 14 and 16 bit with the differences in dither out of the equation.
In actual usage, we decided to use an Apogee A>D due to it's (at the time) superior sounding dither and well trimmed converters, so indeed, we used dither in the application of the technology. This did indeed give much better sub-lowest bit resolution than without dither.
Howard Hoyt
hhoyt, it is my personal opinion that 'some' relatable background material is lost with finite digitization. Both from sampling rate, and available bits at these low level signals.
Apparently, our brain utilizes this low level background material to create the reality that existed when the source was recorded.
If that were not so, I suspect that we would have given up on analog tape and vinyl records, long ago. Dithering may help, or at least cover up major errors like the generation of higher order distortion, but it is not the same material that was there in the first place. If, for example, you see a really HIGH RESOLUTION color photograph, it is uncanny in its realism, yet it is almost impossible to know why, as it is still just a 2 dimensional picture. I think it is the same with audio.
Apparently, our brain utilizes this low level background material to create the reality that existed when the source was recorded.
If that were not so, I suspect that we would have given up on analog tape and vinyl records, long ago. Dithering may help, or at least cover up major errors like the generation of higher order distortion, but it is not the same material that was there in the first place. If, for example, you see a really HIGH RESOLUTION color photograph, it is uncanny in its realism, yet it is almost impossible to know why, as it is still just a 2 dimensional picture. I think it is the same with audio.
- Status
- Not open for further replies.