-
I thought at after 21 or 22 bits anything smaller was just noise. 32 bits sounds like fairy tales. Or has analog technology advanced to that point while I wasn't looking?
https://fscdn.rohm.com/en/products/databook/datasheet/ic/audio_video/audio_processor/bd34301ekv-e.pdf -
130 dB is about 22 ENOB.
32 bits probably refers to the DSP that is in front of the actual DAC.
-
I suspect it just means it can do some of the on-board DSP on 32-bit data.
The SNR is 130dB - equivalent to the LSB of approx. 21 or 22 bits. -
"Supports 16-bit to 32-bit data formats"
So you can feed it 16, 24, or 32 bit audio. The output will be essentially 22 bits due to SNR.
-
According to this article:
https://www.androidauthority.com/why-you-dont-want-that-32-bit-dac-667621/Quote... 32-bit data and 192kHz sample rates have notable benefits in the studio, but the same rules don’t apply for playback. ...
It also mentions that the LG V10, a phone released in 2015, has a 32-bit DAC.
For more perspective also check out the comments made by users at the end of the article.
-
I thought at after 21 or 22 bits anything smaller was just noise. 32 bits sounds like fairy tales. Or has analog technology advanced to that point while I wasn't looking?
Coupling a DAC >24-bits to your ears is pointless. And borders on retarded if the user intends to use it at high volume with ear buds.
-
Give the marking team time and they'll push the engineers to create even higher numbers.
If the goal wasn't audio output.. I'm curious if there could be other possible applications. -
For end consumers, there is little reason to transport audio at more than 16 bit and to replay at more than 24 bit, even with digital volume control considered.
Isn't it that at least some modern (lossy) audio codecs are floating point and therefore do not really have a fixed bit depth? -
Say you're playing a 16-bit audio signal on a 24-bit DAC. From one 16-bit sample to the next, say there is a 1 LSB (relative to 16-bit) step change. Could you use your 24-bit DAC to output samples at 256x the 16-bit clock rate and linearly (or otherwise) interpolate between 16-bit samples to ease the post DAC filtering requirements? Is that what oversampling is?
-
Say you're playing a 16-bit audio signal on a 24-bit DAC. From one 16-bit sample to the next, say there is a 1 LSB (relative to 16-bit) step change. Could you use your 24-bit DAC to output samples at 256x the 16-bit clock rate and linearly (or otherwise) interpolate between 16-bit samples to ease the post DAC filtering requirements? Is that what oversampling is?
I have a sound card that does just that. The Motorola DSP chip does the voodoo by taking what ever the computer sends it and over-samples the life out of it. 20 year old sound card. The company has gone out of business so I can't quickly find more tech info. Found one for sale tho.
https://www.ebay.com.au/itm/Echo-Gina-PCI-Audio-Interface-Sound-Card-24-bit-96khz-Echo-Audio-/254387533631?mkevt=1&mkcid=1&mkrid=705-53470-19255-0&campid=5337590777&customid=&toolid=10001
A [slightly] newer card, but I believe it's similar.
https://www.extremetech.com/computing/55104-review-echo-audio-indigo-io/2 -
Oversampling is doing the interpolation, but no need to use additional resolution for this. It is enough to have the same, in the extremes even slightly reduced resolution.
AFAIK oversampling was used from the very beginning of the CD in the first phillips player and likely also in the Sony version.
Modern Audio DACs are essentially all a kind of sigma delta DAC and this kind of does faster ouput and filtering anyway. So the modulation and oversampling for the DAC may be kind of combined. Even if the DAC externally looks like 24 Bits, it may internally do 8 Bits with massive oversampling and some filtering. -
Echo advertised oversampling rates like 64x or 128x which suggests they meant the internal oversampling of the delta-sigma DAC, which as Kleinstein remarked, is common for all audio DAC ICs produced in this century.
The DSP enabled arbitrary hardware mixing of 8 stereo / 16 mono streams into the two streams fed to the stereo DAC. Perhaps also upsampling them to a common format, but I'm not sure of that. SoundBlasters of the era had similar capability using Creative's proprietary chip, and these were known to upsample the streams to 48kHz causing no end of (perhaps justified) butthurt. SBs also could run a host of effects on their DSP, but CPUs are fast enough that such offload became a nonissue.
Decade earlier, there were standard R2R audio DACs which oversampled to 88 or 176kHz to ease up the demands on hardware reconstruction filters and move part of the filtering to DSP. Recently one vendor revived the concept using Analog Devices R2R industrial DACs and I'm pretty sure you can still buy megabuck gear with discrete R2R.
32 bit is of course marketing wank and parting audiophools from their money. -
It's just marketing wank, more bits = better. BD34301EKV $94 and the EVAL board is $818 at Digi-Key. But finding 32-bit recordings?
I found it extremely hard to design mixed signal 24-bit DAC that actually meets it's performance specs.
I tried ESS parts and they were terrible, they have a cult following and hype, and high price etc. but a simple listening test is hilarious. A cymbal sounding like syran wrap lol interpolate that.
DAC IC's have auto-mute which mutes the output with a zero stream, it's blatant cheating to get "130dB S/N" etc. and I laugh at the hiss and noise jump when that switches off. -
Yeah you can just crank up the number of bits on a sigma delta converter as far as you want, but weather those bits are actually any use is another question.
In practice 16bit is all you need when you have analog volume control after it, but when you do digital volume control then the 24bit is a good idea. But for DSP it makes sense to use 32bit since that tends to be the native number of bits for a lot of processors. The extra bits come in useful when audio is mixed or filtered, preventing clipping and overflows from being problematic.
As far as 32bit float audio, it is never used on a hardware level because floating point numbers are a pain to work with. It is also rarely used in DSP because most DSPs are significantly faster with fixed point math. All the common audio formats also use fixed point. Pretty much the only common use for floating point audio is in audio editing and music creation software. They enjoy not having to worry about clipping at all along the pipeline while x86 PCs are practically just as fast with floating point math as they are with fixed point math, so there is no performance penalty to doing it. -
Oversampling is doing the interpolation, but no need to use additional resolution for this. It is enough to have the same, in the extremes even slightly reduced resolution.
AFAIK oversampling was used from the very beginning of the CD in the first phillips player and likely also in the Sony version.
Actually, that was the first divergence. Philips felt that they could only achieve 14 bits with decent linearity at launch, they introduced 4x oversampling as a form of compensation for this. Sony launched with 16 bits and no oversampling. The Philips players were generally regarded to sound much better as the oversampling pushed the required post-DAC analogue filtering well above the audio frequency range. Sony had to use a brick-wall analogue filter which caused lots of audible phase and amplitude errors in the audio passband.
By the next generation, Philips had perfected decent 16 bit DACs but retained the 4x oversampling for the more relaxed filtering requirement. -
In the early days of sound blaster for the PC, there was DOS project that aimed to convert 8-bit to 16-bit.
In the period before the 16-bit soundblaster, apparently people had recorded stuff in 8-bit.
Whilst it couldn't do much about the sample source being recorded in 22050Hz, the 8-bit data was 'dithered' to 16-bit. What that means is they rounded the sharp, jagged edges using alot of floating point. You had to have a 486 DX if memory serves.
I remember giving it a go once. I recorded a song in 8-bit mode and ran it through the software. I think it took about 25 minutes for a 3 minute song. It sounded not as harsh as the 8-bit, but it wasn't great either.
When expanding out the bit depth, I have to wonder what transform is being used. I hate to use this analogy but it's like when they tried to colorize b/w movies. Sometimes, you just shouldn't. -
Oversampling is doing the interpolation, but no need to use additional resolution for this. It is enough to have the same, in the extremes even slightly reduced resolution.
AFAIK oversampling was used from the very beginning of the CD in the first phillips player and likely also in the Sony version.
Actually, that was the first divergence. Philips felt that they could only achieve 14 bits with decent linearity at launch, they introduced 4x oversampling as a form of compensation for this. Sony launched with 16 bits and no oversampling. The Philips players were generally regarded to sound much better as the oversampling pushed the required post-DAC analogue filtering well above the audio frequency range. Sony had to use a brick-wall analogue filter which caused lots of audible phase and amplitude errors in the audio passband.
By the next generation, Philips had perfected decent 16 bit DACs but retained the 4x oversampling for the more relaxed filtering requirement.
A bit later, before the 1-bit DACs, a few cheap designs came out with a DAC running at 88.2kHz. Some switching logic and a 4066 on the output side. Because two DACs was too expensive. I've got an old schematic here somewhere. I'll have to upload it sometime. You look at the output section and go
-
A bit later, before the 1-bit DACs, a few cheap designs came out with a DAC running at 88.2kHz. Some switching logic and a 4066 on the output side. Because two DACs was too expensive. I've got an old schematic here somewhere. I'll have to upload it sometime. You look at the output section and go
Many of the cheaper ones still do this for the DAC, using an analogue switch and a sample and hold to generate the stereo output, as in general the DAC values are close enough at high frequency to not cause much slew, and the crosstalk is not too bad as a result. You can still hear it for audio coming out of one channel only as a faint high frequency signal in the other quiet channel. -
Couple points:
* If you're dealing with 32-bit samples, a true 32-bit DAC is likely going to yield a better SNR than a 24-bit fed with samples down-quantized from 32-bit to 24-bit;
* Sigma-delta DACs are really all digital. And to achieve a given SNR, it can be cheaper to design a 32-bit SD DAC than a 24-bit one, for instance.
Of course, if the question is: can you get close to the ideal SNR of ~192 dB, the answer is no, but that would be missing the two above points - they are not made for this.
Now sure there can be some marketing - which allows to sell those 32-bit DACs at a higher price point than 24-bit DACs, whereas as I said above, they can be actually easier to design for a given target SNR, and with possibly simpler digital filters needed, even possibly take actually a bit less silicon area...
-
Modern Audio DACs are essentially all a kind of sigma delta DAC and this kind of does faster ouput and filtering anyway. So the modulation and oversampling for the DAC may be kind of combined. Even if the DAC externally looks like 24 Bits, it may internally do 8 Bits with massive oversampling and some filtering.
Analog Devices published an excellent document describing the three basic architectures, single-bit, multi-bit, and MASH (multistage noise shaping):
https://www.analog.com/media/en/training-seminars/tutorials/MT-023.pdf
Briefly, single-bit delta-sigma converters have higher linearity but require extra modulator stages to suppress idle tones and this makes them susceptible to instability from overload. Multi-bit delta-sigma converters can have fewer modulator stages, but DAC linearity limits linearity.* If you're dealing with 32-bit samples, a true 32-bit DAC is likely going to yield a better SNR than a 24-bit fed with samples down-quantized from 32-bit to 24-bit;
Do you have any examples where that is the case and it is not just marketing? TI's (Burr-Brown) highest SNR audio ADCs and DACs are 24 bits and have a higher claimed SNR than the "32-bit" converters from AKM.
Some instrumentation converters claim slightly more than 24 bits of noise free resolution.
-
* If you're dealing with 32-bit samples, a true 32-bit DAC is likely going to yield a better SNR than a 24-bit fed with samples down-quantized from 32-bit to 24-bit;
Do you have any examples where that is the case and it is not just marketing? TI's (Burr-Brown) highest SNR audio ADCs and DACs are 24 bits and have a higher claimed SNR than the "32-bit" converters from AKM.
To make sure my point is fully understood:
1. I am assuming a 32-bit DAC and a 24-bit DAC with the same SNR.
2. I am explicitly talking about dealing with 32-bit samples, which, if it was not clear enough, means samples that are natively generated as 32-bit numbers.
3. If you have 32-bit samples, and feeds a 24-bit DAC, you have to down-quantize. Merely truncating creates truncation distortion which is rather nasty. The usual way of mitigating this is to *dither*, which basically is adding some random noise to the signal. It's a trade-off. Usually less annoying than pure distortion, but added noise nonetheless. Then end-result will be noisier than the 32-bit samples directly fed to a 32-bit DAC with the same SNR.
4. Of course, if you're dealing with 24-bit samples, using a 32-bit DAC wouldn't make sense. Actually, you'd get worse results too since you'd have to up-quantize here, which would inevitably make the signal noisier, even if just a little.
5. All in all, it's best to use a DAC with the same resolution as the samples you're feeding it with. Any change of quantization WILL add noise.
6. Some may question the benefits of 32-bit samples to begin with. And if you're considering recorded music played on an audio player, you'd be right: there is no native 32-bit material available. But there still are use cases for this. For instance, digital mixing consoles, or any device that does DSP with a native width larger than 24 bits.
-
1. I am assuming a 32-bit DAC and a 24-bit DAC with the same SNR.
2. I am explicitly talking about dealing with 32-bit samples, which, if it was not clear enough, means samples that are natively generated as 32-bit numbers.
That is only true for a very specific meaning of "natively generated as 32 bit numbers" Most audio you will find in any format has been subjected to some combination of signal processing and then truncated to a target output resolution. Taking 32 bit audio produced this way, then truncating them to 24 bits is likely going to produce equivalent results to if the original producer had simply truncated the original to 24 bits.Quote3. If you have 32-bit samples, and feeds a 24-bit DAC, you have to down-quantize. Merely truncating creates truncation distortion which is rather nasty. The usual way of mitigating this is to *dither*, which basically is adding some random noise to the signal. It's a trade-off. Usually less annoying than pure distortion, but added noise nonetheless. Then end-result will be noisier than the 32-bit samples directly fed to a 32-bit DAC with the same SNR.
That depends on your DAC. If the 32 bit DAC doesn't have any better linearity (especially DNL) than the 24 bit DAC that is going to set your distortion floor and dithering is not really going to fix it. And if the SNR of the 24 bit DAC is < 24 bits (say 20 bits), then you can easily add enough dither to decorrelate the quantization noise without materially affecting the overall SNR. Technically it still adds noise, but the added noise can be inconsequential, and in any cases it is exactly the same noise you would have to add to generate native 24 bit audio in the first place.Quote4. Of course, if you're dealing with 24-bit samples, using a 32-bit DAC wouldn't make sense. Actually, you'd get worse results too since you'd have to up-quantize here, which would inevitably make the signal noisier, even if just a little.
An ideal 32 bit DAC with the lowest 8 bits zeroed is an ideal 24 bit DAC. If you have a 32-bit DAC and 24 bit audio dithering and up-quantizing might improve the distortion products for very particular waveforms at the expense of noise, but not re-quantizing is just as good as the 24 bit DAC would be anyway.Quote5. All in all, it's best to use a DAC with the same resolution as the samples you're feeding it with. Any change of quantization WILL add noise.
But completely impractical and not important at all in practice. Sample depth conversions are easy, common, and have insignificant impact on SNR when done properly. At the same time it is very useful to have intermediate formats that are 24-bit, 32-bit, or floating point, but since not all sources are the same so it is mandatory for mixers to combine them. For a final product 16 bits is about all the dynamic range you can use anyway, so carrying around an extra 8 or 16 bits is just a waste, but again it can be useful to re-quantize to 24 bit or higher for signal processing at the playback end. -
We are talking about the actual hardware that does the DAC. I don't think you can reasonably make those. The exponent factor needs hardware analog multiplier to implement, and making a linear enough multiplier is harder than it seems to be.
What I'm getting at is that in practice, some codecs already "transport" the audio using more than 16 bits (after uncompressing) per sample. How many "integer bits" of precision those floating bit samples have can be a complex question, not to mention how much degradation the lossy compression really introduces.
At least in theory, if the floating point samples do not exactly map to 16 bit integers, mapping them to 24 bit integers means less error. -
The reason why people push 24 bit is for volume control. You see, analog things are expensive, they don't scale with Moore's law. You want a system to be mainly digital, so that they can scale and you get to save money. If your intended SNR is 98dB (CD quality), then you can use the rest 48dB to do volume control, and that is actually a large range.
It's more that a good stereo pot like an ALPS costs $40-$50 (maybe less in 1ku volume), while attenuation in firmware costs 2kB of code space.
-
Many DACs also have PGAs built in, so just a little bit of code to write to the register(s) that controls the PGA.