I don' know you DAC, but all DAC's behavior is the same, if to send to their input less bits than they can adopt.
Suppose you take DAC with 32bit data register, which have only 17 bit ENOB (for example - PCM5102A, -105dB THD+N) - there will be no difference if you send 17bit or 24 bit signal to it's input, in any case 15 LSB of 32 bit data are not relevant.
Now you attenuate the signal at -24dB -> shift 4 bits right. How many bits this DAC will use to conversion? 17-4=13
Let take DAC with ENOB=20, 20-4=16 -> this is good, because most of the music signal are 16 bit.
Alex.
Suppose you take DAC with 32bit data register, which have only 17 bit ENOB (for example - PCM5102A, -105dB THD+N) - there will be no difference if you send 17bit or 24 bit signal to it's input, in any case 15 LSB of 32 bit data are not relevant.
Now you attenuate the signal at -24dB -> shift 4 bits right. How many bits this DAC will use to conversion? 17-4=13
Let take DAC with ENOB=20, 20-4=16 -> this is good, because most of the music signal are 16 bit.
Alex.
Usually maximum possible ENOB is interesting.
You can use calculator:
https://www.analog.com/en/design-ce...gn-tools/data-conversion-calculator.html#ENOB
You formula is also correct.
Alex.
THD + N -81.7 dB(A)
Noise floor -104.5 dB(A)
How does that fit with your model of DAC behaviour?
This calculator does not take into account the reduction in amplitude of the DAC output signal when measuring SINAD, which makes it difficult to find ENOB that is tied to the entire measurement scale.
Here they write like this.
https://www.analog.com/media/en/training-seminars/tutorials/MT-003.pdf
I thought carefully and realized that this formula is not correct, since when the output signal decreases, the measured SINAD will decrease and at the same time ENOB itself should not decrease. According to the formula that I gave earlier, such a condition cannot be met, since the correction term will be negative when the signal becomes less than full scale. In that formula, you need to swap the output voltage with full scale.
I'm not sure I'm understanding you here as you seem to be agreeing with me--did you mean increase rather than decrease?
My belief that THD rises with output level is based on DAC data sheets and measurements of complete DACs. In most of these, I see THD+N graphs with a smooth decline from 0dB at the noise floor until a valley somewhere between -12 through -6dBFS, where THD+N flattens then rises, indicating distortion (or correlated noise) is above the noise floor only in the top two bits. I certainly could be wrong as FFT spectra are rarely shown anywhere but 0dB and -60dB output levels. -60dB spectra are usually clear of distortion down to the FFT noise floor (between -120 & -160dB).
Many audio ADC/DAC have THD rising near 0dBFs, but this is not because of "bits'".
What I mean is that higher digital attenuation will result in higher THD because there are fewer bits involved in the conversion.
What do you mean, Marcel?
"Noise Floor"? It is just a visual representation, this is not SNR.
With exactly the same signal you can obtain very different Noise Floor, changing averaging or/and FFT size.
I did an experiment once with one of the D-class amplifiers. Just turned it on, put some music and set comfortable volume in windows mixer, volume at amp side was at 0dB
Then just took the scope and looked at I2S signal: only 12-13 bits were toggling.
It is not so dramatic for DAC however, since there it does not apply additional gain (amplifier gives +20dB).
One experiment that I still would like to do is to check whether digital attenuation in TAS5548 works better. Internal data buses are 32 bits, but that's all from the datasheet. So really depends on their SDM design.
Best Regards,
Vladislav.
Exactly, an increase in harmonics with an increase in the signal to 0dB is not always associated with a decrease in DAC bit depth. However, for some reason, you interpret such an event as a reduction in DAC bit depth and then proceed to reduce the number of bits in ENOB to decrease volume. In my view, doing so is hasty; this arithmetic does not accurately reflect the real situation.
I conducted measurements of THD+N at different levels of digital volume in the DAC and calculated ENOB using two formulas, one without normalization and one with normalization.
As evident from the measurement results and calculations, the ENOB full scale remains relatively stable when reducing the volume. Looking at the ENOB full scale, you can observe that the bit depth decreases due to the characteristics of the DAC power stage, not due to a reduction in bit depth during volume adjustment. This is something you also mentioned, but you seem reluctant to consider it in your reasoning. When discussing the impact of digital volume on ENOB, simply discarding bits from ENOB is futile because bits are discarded at the input of the DAC. If the DAC has a 24-bit input, reducing the volume involves subtracting bits from the 24-bit stream, not from the ENOB.
The last column does not make any sense to me as it can be considered as cheating 
ENOB is only defined for full-scale signal.
ENOB is only defined for full-scale signal.That is true, digital attenuation has nothing to do with ENOB of the DAC. One should only consider SINAD/THD+N.
Sure.The last column does not make any sense to me as it can be considered as cheating
Alex.
The last column clearly shows that the decrease in ENOB is not due to a reduction in the bit depth of the signal. This fact illustrates that predicting a decline in ENOB by subtracting bits from ENOB when reducing the volume is meaningless.The last column does not make any sense to me as it can be considered as cheating
Agree.
For me, ENOB does not accurately reflect the real situation at the output of the DAC. In the case of powerful PWM DACs, the forefront is occupied by distortions introduced by PWM amplifiers and low-pass filters, rather than the bit depth of the digital signal generating this PWM. ENOB implies that the DAC cannot reproduce bits below ENOB, but in reality, this is not the case. The DAC can output a signal below the bit depth of ENOB calculated at 0dB. Therefore, in my view, it is pointless to use the ENOB parameter to predict the capabilities of the DAC at lower volumes. It is much more important to know the level of noise and distortion generated by the DAC at various power levels. In simpler terms, it is more important to see the DAC's level sweep than to look at the ENOB obtained at 0dB. It's worth noting that in my case, 0dB corresponds to 200W at 4 ohms, and having a THD+N of 0.008% at such power is an acceptable value. The amplitude of the sinusoidal current consumed by the load is 10A, and the power output of the DAC maintains a THD+N of 0.008% at such currents, which is very good for such levels currents. The fact that ENOB is calculated as 13 bits does not truly reflect the real situation.
The situation is really strange - we making a DAC with 0.0000x%, then sending the signal to the amplifier, which has 0.00x-0.x%, and then to the speakers, that have x% (well, 0.x%, at best case..).
(Not in this design, but “in general”).
Alex.
Presumably you know that THD+N has been considered a useless metric for the degree of distastefulness of an audio device since at least 1938?
Dr. Sean Olive, and Dr. Earl Geddes have both made that quite clear, each in their own way.
Understood, its a convenient metric since it rolls everything up into a single number. But if the number is meaningless, then why use it?
On the subjective side, how does this dac/amplifier sound?
This DAC, when operating across the entire frequency range, sounds more neutral than colored, with sufficiently high sound detail; there is some sound coloring, but it is very minor. The DAC copes very well with a speaker impedance drop down to 1 Ohm, which sometimes allows for the realization of the potential of multi-way speakers with complex input impedance. Despite being lower in power and with slightly better THD+N performance, another DAC underwent listening tests, even in studio conditions. Sound engineers appreciated its neutrality and detail, choosing to keep it permanently. However, this is all when the DAC operates across the entire frequency range.
When this DAC is used in active three-way speakers, it is barely audible; one can perceive changes in tonal balance or sound character at different settings of the frequency divider, or with different corrective EQs, or various filters in the digital crossover, but the DAC's character is imperceptible. There are already a couple of three-way active speakers built around a 50W DAC; one speaker is working in a recording studio, and the person is very satisfied with the result and the flexibility of the speakers in adapting to their needs. I have assembled a near-field three-way active speaker system. You can hear the drivers, you can hear the speaker blending, but you can't hear the DAC.
Yes, that's right. There isn't one universal amplifier or speaker system that can meet all expectations. I mentioned the recording studio only to illustrate that this DAC has been auditioned not only in domestic settings but also in studio environments.
I think it's time to show the implementation of my idea in hardware.
Module dimensions in millimeters.
Power consumption of each module at idle is 14W.
The level of voltage-weighted noise at the amplifier output is no more than 300μV.
There are no clicks when turning the power on and off at the amplifier output; modules can be turned on and off in any sequence.
The module has protection against overheating and excessive current consumption; also, the absence of DC current is monitored at the amplifier output, with a protection response time of 150 milliseconds.
If anyone is interested in applying these modules to their active speakers, feel free to ask questions, and I'll respond.
Module dimensions in millimeters.
Power consumption of each module at idle is 14W.
The level of voltage-weighted noise at the amplifier output is no more than 300μV.
There are no clicks when turning the power on and off at the amplifier output; modules can be turned on and off in any sequence.
The module has protection against overheating and excessive current consumption; also, the absence of DC current is monitored at the amplifier output, with a protection response time of 150 milliseconds.
If anyone is interested in applying these modules to their active speakers, feel free to ask questions, and I'll respond.
Last edited: