No. It is because they actually need to do DSP calculations and need the extra precision for the processing operations. You don't need it for a non-DSP chain.
hifi-advice.com - Upsampling - DCS Review and explanation
bear in mind that this review is biased simply because DCS is in a business of selling upsampling hardware
Nick
bear in mind that this review is biased simply because DCS is in a business of selling upsampling hardware
Nick
Last edited:
And going 64-bit floating point is worse. Instead of just adding useless zeroes, you actually occur rounding error in converting to floating point (and then back to integer again for the DAC). Totally useless unless you need the extra room for DSP operations.
Back from theory to real world, no DAC is perfect. You assumptions are valid if we stay in the field of pure maths. Real world measurements show that cleanness of DAC output spectrum is better with upsampled 24bit signal than with original 16 bit data (96/24 better than 48/16, e.g.) though no useful information has been added.
That's correct, but same applies for anyone's involved in the business reasoning. Class D power amplifier designer would explain you that it is the best way how to make a linear analog amplifier
It is very easy to look below the surface.
Yes, quite a lot, and also posted in another threads here. Frankly, no reason to get involved in another never ending web discussion.
And you are right, depends on DAC used. But none of them is perfect and they almost always profit from zero padding and insertion of data with smaller LSB steps.
If a DAC is 24bit and samples have only 16 bits, zeroes have to be padded in any case. Either by the playback SW chain (something you can influence), or the driver (accepting various formats, outputting only a single or two supported by the HW natively), or the soundcard controller (talking to the DAC chip), or the DAC chip itself (if it accepts various formats of I2S)
I very much doubt padding zeroes in the software chain (without resampling) has any effect whatsoever - they would have to be padded somewhere further down the stream anyway.
I very much doubt padding zeroes in the software chain (without resampling) has any effect whatsoever - they would have to be padded somewhere further down the stream anyway.
And just as has been posted in a number of threads here, zero-padding mostly has no effect, and in some cases a negative effect - no reason to get involved in another never ending web discussion.
Remember that you are not creating any "smaller LSB steps" - the steps stay the same, unless you do non-bit-perfect DSP operations.
The problem with your original statement was that you made no difference between upsampling and zero-padding - two completely different operations.
Misunderstanding. When you upsample from 48/16 to 96/24, new 24 bit samples are inserted (in place of added zeros). They do not bring any new information, but they do bring smoother steps than original 16bit LSB, speaking about newly inserted samples. This is beyond any discussion.
Last edited:
Can we please separate the two completely different operations of upsampling from 48 to 96 kHz sample rate - a non-bit-perfect, transformatory DSP operation, and word-extending the data from 16 to 24 bits?
Merely word-extending the data from 16 to 24 bits does not create any more information or precision - the resolution of the data is still 16 bits, but you have more empty bits to shift around. Upsampling from 48 to 96 kHz requires recalculating the data, and 24 bits gives more precision *for that calculation*. It still doesn't change the fact that the original material only has a 16-bit SNR - the SNR (and resolution of the original data) doesn't improve.
Please don't use the expresson "smoother steps", that just reflects the common misconception that a digital signal consists of "steps".
This explain very well the improvements...
There is not about here to be added (new) informations. Adding informations to the original ones is just wrong, and it may not be happen. Then the result it may be a different file than original it was.
But adding just "neutral" informations, which it not alter the original one, and it help the DAC processing to get smother, or improve the approximation of the analogue signal to be reconstituted, this it can only improve the final result.
No matter the theoretical considerations bring it in by some, the perceptual improvements of an upsampling process remain just a fact.
As I could see, there are some companies which produce upsampling dedicated devices. They may not be so foolish to invest in something just because such processing it may be seen as useless by someone...
No. It can also worsen the result, by causing extra load from haing to move the empty zero-filled bytes in and out.
Only if "fact" is defined very loosely.
There are also companies that produce magic boxes that improve sound quality by "neutralizing" electric fields in your listening room...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.