Re: List of digital consoles
Assuming the implementation of the processing (fixed or floating) is done competently, the larger determinant of sound quality is due to the converters and their clocking. Just about all modern audio ADC's and DAC's spit out or take in 24 bits (which is fixed point BTW), but none of them give anywhere near 24 bit performance. This seems to be an overlooked spec on digital boards. Cheap converters are about 90 dB dynamic range (DNR), and -85 dB THD. At 6 dB per bit (actually 6.02), this corresponds to 15 bits! Nice converters are about 107 dB DNR and -100 dB THD, for nearly 18 bit performance. The very best converters, of which there are few models available, are 123 dB DNR and -110 dB THD, which corresponds to 20.5 bits of DNR. I doubt these 123 dB converters are finding their way into the 'mass market' boards.
And if these converters are not clocked with local crystal oscillators of careful design, their distortion performance will be degraded further. This means there are no PLL's in the chain. Using outboard clocks will mean a PLL will be used inside the board to generate the master converter clock.
About 15 to 20 years ago, the dominant DSP in use for digital audio was the fixed point Motorola (now Freescale) 56K, which is 24 bits, with a double precision 48 bit mode. Yamaha had their own in-house DSP design, which I believe was/is fixed point. The floating point DSP's of that era were much more expensive, and required more memory to hang off them, further increasing the cost. But the only real (but small) limit on the signal quality of 24 bit fixed point was for use on IIR filters, particularly very high Q peaks or notch settings. This is where using the 48 bit double precision mode could help, but I doubt for normal use in a live system you could even tell the difference. When I did the experiment back then, switching between 24 and 48 bit mode, it was hard to hear in a studio control room, and then only on some signals with higher Q filter settings. As the double precision operations would take about 8 times as long and twice the memory, they are not used for the entire processing chain.
Today, floating point DSPs have really come down in price, and they have enough internal memory, so that any new design would be foolish to not use a floating point DSP. It is interesting to note that a 32 bit floating point word still only has a 24 bit mantissa, or fractional part. 8 bits are used for the exponent, which gives the programmer freedom from saturation (clipping) considerations. But this floating point word has to be converted back to 24 bit fixed point for output to the DAC. Also note that there are DSPs (non programmable ones) inside the ADCs and DACs which perform FIR filtering with fixed point arithmetic!
Assuming the implementation of the processing (fixed or floating) is done competently, the larger determinant of sound quality is due to the converters and their clocking. Just about all modern audio ADC's and DAC's spit out or take in 24 bits (which is fixed point BTW), but none of them give anywhere near 24 bit performance. This seems to be an overlooked spec on digital boards. Cheap converters are about 90 dB dynamic range (DNR), and -85 dB THD. At 6 dB per bit (actually 6.02), this corresponds to 15 bits! Nice converters are about 107 dB DNR and -100 dB THD, for nearly 18 bit performance. The very best converters, of which there are few models available, are 123 dB DNR and -110 dB THD, which corresponds to 20.5 bits of DNR. I doubt these 123 dB converters are finding their way into the 'mass market' boards.
And if these converters are not clocked with local crystal oscillators of careful design, their distortion performance will be degraded further. This means there are no PLL's in the chain. Using outboard clocks will mean a PLL will be used inside the board to generate the master converter clock.
About 15 to 20 years ago, the dominant DSP in use for digital audio was the fixed point Motorola (now Freescale) 56K, which is 24 bits, with a double precision 48 bit mode. Yamaha had their own in-house DSP design, which I believe was/is fixed point. The floating point DSP's of that era were much more expensive, and required more memory to hang off them, further increasing the cost. But the only real (but small) limit on the signal quality of 24 bit fixed point was for use on IIR filters, particularly very high Q peaks or notch settings. This is where using the 48 bit double precision mode could help, but I doubt for normal use in a live system you could even tell the difference. When I did the experiment back then, switching between 24 and 48 bit mode, it was hard to hear in a studio control room, and then only on some signals with higher Q filter settings. As the double precision operations would take about 8 times as long and twice the memory, they are not used for the entire processing chain.
Today, floating point DSPs have really come down in price, and they have enough internal memory, so that any new design would be foolish to not use a floating point DSP. It is interesting to note that a 32 bit floating point word still only has a 24 bit mantissa, or fractional part. 8 bits are used for the exponent, which gives the programmer freedom from saturation (clipping) considerations. But this floating point word has to be converted back to 24 bit fixed point for output to the DAC. Also note that there are DSPs (non programmable ones) inside the ADCs and DACs which perform FIR filtering with fixed point arithmetic!
