Theoretically 0 %. There are quantization errors, but with proper dithering and a perfect DAC, that is counted as noise rather than distortion.
H
HAYK
When sampling, the error of the output is +/-0.5LSB, forcibly there is distortion even if the DAC is perfect.
The lowest I found in datasheets is that of PCM1704, it says 0.9% with 16bit data, when it is 0.6% with 24 bit.
A very linear 16bit DAC the LTC 1597 has a distortion lower than the 24 bit 1704, I wonder what it is at -60db.
The lowest I found in datasheets is that of PCM1704, it says 0.9% with 16bit data, when it is 0.6% with 24 bit.
A very linear 16bit DAC the LTC 1597 has a distortion lower than the 24 bit 1704, I wonder what it is at -60db.
Last edited by a moderator:
You asked about total harmonic distortion, not THD+N. The error can be made noise-like with proper dithering, then it is not part of the harmonic distortion.
16-bits is not very high resolution. A 1-bit DAC can do 16-bits with essentially perfect linearity.
Ed
Ed
I think I remember from college SINAID for a digital representation is 6db x #bits, so 16 is 96 referenced to full scale ideal. Looks like the pcm1704 is close to perfect. At -60db signal level, SINAID will be 36db down from the signal. A 24 can ideally do 144db, but most only achieve somewhere between 18-19 bits.
H
HAYK
This what I also thought. -60db is +/-32 levels that is 6bit definition, so the minimum THD should be -36db. The 1704 shows -42db, this what made me ask the question.
You are still mixing up THD+N with THD.
Regarding undithered THD+N with a perfect DAC, you get 1.76 dB extra because of the differences in probability distribution between sine waves and quantization errors. Besides, you get some decibels extra when you don't measure over the full Nyquist bandwidth, but something smaller.
With triangular dithering, you lose 4.77 dB, but the quantization error then behaves very similarly to additive noise.
Regarding undithered THD+N with a perfect DAC, you get 1.76 dB extra because of the differences in probability distribution between sine waves and quantization errors. Besides, you get some decibels extra when you don't measure over the full Nyquist bandwidth, but something smaller.
With triangular dithering, you lose 4.77 dB, but the quantization error then behaves very similarly to additive noise.
Last edited:
From some 1kHz step level measurements I have made, THD is btn 0.28% and 0.7% at -60dB level.
Noise is btn 0.5% and 4%
All these DACs are based on TDA1387 (Abraxalito's implementations)
George
Attachments
Last edited:
H
HAYK
The measurement of 1704 are with fs 768khz. I suppose with 44.2khz it will be lower as the glitches and settling time will act less . How low it can go?
Does the 20bit 1702 will be similar?
The main trouble at low level, I read, is the MSB calibration. I wonder if the $0.5 TDA1387 with 2.2% THD can get bellow 1% if I load only the magnitude and let the sign bit, MSB, switch the polarity.
Does the 20bit 1702 will be similar?
The main trouble at low level, I read, is the MSB calibration. I wonder if the $0.5 TDA1387 with 2.2% THD can get bellow 1% if I load only the magnitude and let the sign bit, MSB, switch the polarity.
H
HAYK
Thank you very much, this is very interesting. Abraxalito has many versions, is this with 2 DAC in differential or multiple in parallel?
H
HAYK
To compare apples to apples, the measurement of pcm1704 probably is similar to 1702 shown bellow is filtered at 30khz and not dithered. Whereas the tda1387 is typ. 1.7% THD+N with 80khz maybe 1.5% with 30khz. This meanes only -6db decrease can be hoped from modding the TDA.
What I find sad that paralleled DACs measured by George doesn't decrease the noise. Normally, every doubling the number of DACs, the noise should decrease by 3db.
What I find sad that paralleled DACs measured by George doesn't decrease the noise. Normally, every doubling the number of DACs, the noise should decrease by 3db.
Attachments
That is true only if the noise is uncorrelated. Digitally-generated noise will be deterministic if not seeded from a truly random source.HAYK said:
Ed
H
HAYK
What is wrong about George's measurements, he uses dithering option in his generator. This input noise is decreasing the THD for sure as Marcel explained but increasing considerably the overall noise. As the dithering noise is in the recorded data, it is useless to go low noise audio DAC. This means paralleling DACs doesn't pay specially if your estimation of noise digital nature, doesn't decrease.
The 1 bit TDA1305 THD+N is measured by 18bit data and A weighted, the DS says.
I think, if I adjust the MSB of the TDA1387 as PCM56 or AD1855 can, it will be as good as the best 16 bits are.
The 1 bit TDA1305 THD+N is measured by 18bit data and A weighted, the DS says.
I think, if I adjust the MSB of the TDA1387 as PCM56 or AD1855 can, it will be as good as the best 16 bits are.
Attachments
I'd say George's measurements are fine while those of TI are not. Digital audio recordings should be dithered, because dithered quantization sounds far better than undithered quantization.
H
HAYK
If 24bit data is used to supply the 16bit DAC, the dithered noise is +/-LSB of the 24bit. The 16bit DAC when truncate the lower bits it's LSB is not dithered. Does this mean that listening 24bit recording on 16bit DAC sound bad?
If I find George's measurements wrong because it doesn't compare apples to apple but what in real world one will get. By this we can't know by how much better he got by using 32// DACs As I mentioned, according to his measurements, there is no any need to look for low noise DACs.
One thing comes to my mind, the TDA1540 14bit DAC. it truncates the 2 lower bits and by delta sigma it integrates into active ones averaged. Does this operation cleanes the dithered noise?
If I find George's measurements wrong because it doesn't compare apples to apple but what in real world one will get. By this we can't know by how much better he got by using 32// DACs As I mentioned, according to his measurements, there is no any need to look for low noise DACs.
One thing comes to my mind, the TDA1540 14bit DAC. it truncates the 2 lower bits and by delta sigma it integrates into active ones averaged. Does this operation cleanes the dithered noise?