TI says:
"8.2.1.2 Detailed Design Procedure
The design of the application circuit is very important in order to actually realize the high S/N ratio of which the PCM1795 device is capable, because noise and distortion that are generated in an application circuit are not negligible."
At least they acknowledge the chip will be implemented on an imperfect PCB and powered by an imperfect power supply.
Nick
This really deserves a different thread in a different forum, though I don't see a "Digital Control Voltage Source" subforum. I'm on that synth site with that pic of the two cats, maybe the DIY section there would be good. This might fit under the "Instruments and Amps" forum here, but it would be nice to have an "Electronic Instruments" subforum off that.
But to your question, it appears what you're doing will pretty much work, but ... I see enough 'buts' here to make me think twice. The two-pole filter might be enough to reduce the 490Hz PWM signal to low enough that you don't hear it frequency-modulate the oscillator. The microcontroller power voltage needs to be rock steady, because that's what the average PWM voltage and your final control voltage is based on, though they're always run on decent regulators so that might not be a problem.
Rather than PWM, I'd go with an old-fashioned DAC that has an internal r/2r ladder or similar current-switching scheme, as that will give a direct DC output of the voltage you want (though it will still need to be amplified with an opamp to give 1v/octave over the the full oscillator range), and do so instantly. Yeah, that may need some "port expansion" and who knows what other mods to the Arduino, but that gets beyond the scope of a very marginally on-topic post.
Questionable indeed. Does "32 bits" do at least 24? There are a few reasonable ways to measure, but if you are off by at least 8 bits that is even more problematic.
I have to admit saying this thing is a 32bit DAC is like saying an engine and 6-speed transmission are a "6-speed engine." They wouldn't do that with cars because a they know too many people are knowledgeable enough that they couldn't get away with it.
But in audio, we've had the RMS watt for over 40 years, even in spite of the FTC disclaiming the thing.
But in audio, we've had the RMS watt for over 40 years, even in spite of the FTC disclaiming the thing.
If the DAC accepts 32 bit data then it is a 32 bit DAC. Are we going to get our panties in a knot now about how no 24 bit DAC achieves 24 bit analogue performance either? Are we going to label all of those as fraudulent too?
The point is, the fact that the analogue output on any DAC has a limitation imposed by the laws of physics such that you won't exceed 21-22 bits performance in normal conditions has nothing to do with the processing and data input side of the thing being 32 bits, 24 bits, 48 bits or whatever. Where accuracy of digital processing is required the more bits you've got the better.
The data sheets are aimed at technically knowledgeable people who should know this. Whether or not the general public are being misinformed by any technical sticker plonked on a product or by some marketing department advertising is besides the point. People get the wool pulled over their eyes all the time, especially audiophiles and the marketing departments therein. At least these DACs do accept and process 32 bit data, or do you want some snake oil to polish your silicon chips with too?
The point is, the fact that the analogue output on any DAC has a limitation imposed by the laws of physics such that you won't exceed 21-22 bits performance in normal conditions has nothing to do with the processing and data input side of the thing being 32 bits, 24 bits, 48 bits or whatever. Where accuracy of digital processing is required the more bits you've got the better.
The data sheets are aimed at technically knowledgeable people who should know this. Whether or not the general public are being misinformed by any technical sticker plonked on a product or by some marketing department advertising is besides the point. People get the wool pulled over their eyes all the time, especially audiophiles and the marketing departments therein. At least these DACs do accept and process 32 bit data, or do you want some snake oil to polish your silicon chips with too?
While on this topic, is there an audio interface that can get 32-bit audio to the ESS from a source?
For example, is there a 32-bit 192KHz multichannel USB to I2S interface?
The built-in per-channel digital attenuation might benefit from getting the 32-bits to ESS. Is my understanding correct?
For example, is there a 32-bit 192KHz multichannel USB to I2S interface?
The built-in per-channel digital attenuation might benefit from getting the 32-bits to ESS. Is my understanding correct?
Except its tail isn't a leg and yet the DAC is accepting and processing 32 bit data. Funny that.
It doesn't matter, any actual data beyond the 21-22bits is simply noise. If you feed 24 bit data into the ESS it simply fills up the extra 8 bits with zeros. The important part here though is that if you've got a 24 bit system with 22 bit data, then you attenuate the signal in dB equal to a loss of 4 bits, you're going to throw away 2 bits of data. If you've got a 32 bit system this does not occur as the system has plenty of headroom to cope with the signal attenuation on the digital side.
The DAC doesn't process, it converts. It's the DSP doing the processing.
But then, I remember as a child in the 196os being annoyed at people referring to a battery powered handheld radio as a "transistor."
No a DAC receives a digital signal, processes it into a suitable format and then converts it into an analogue signal. The processing part is as necessary as every other part and all modern DACs contain DSP processing that implements the standard FIR brickwall filters and nowadays a heluva lot more. In other words you cannot have the analogue output without the preceding digital processing stages.
Are we really trying to argue over something like this though?
As far as I remember PCM has such standard (which is more accurate than the usual 2s-complement to calculate the signal amplitude) but may be not commonly implemented.
Exactly, that's why digital attenuation should be avoided as possible, while otherwise loosing on the analog headroom.
32 bit full range
+----------------------------------------------------+
22/21 bit's = analog noise level
+------------------------------------+
Resulting maximal analog headroom
<------------------------------------>
Attenuated digital signal
........+--------------------------------------------+
Resulting attenuated analog headroom
........<---------------------------->
Hp
But that can actually be useful!
Consider that the analogue headroom often exceeds that of the rest of the system, and a properly done digital gain control will track better then even an expensive pot...
Now consider that the conversion between a 32 bit value and a 24 bit one is not actually entirely trivial to get right (particularly for a floating point input) and actually being able to throw 32 bit data at the converter and let it handle the final gain control and conversion is actually something I would buy into if I was designing a processing box provided the analogue specs were competitive.
Converters have an ENOB (or noise floor, or noise power density or equivalent) spec for a reason and they all specify the input word length as the headline number (And have done forever).
Consider that outside the audio world, you can indeed get more then 24 bits of precision, it just requires a measurement bandwidth of a few tens of Hz, and that this is sometimes useful (And it is outside the audio world that defined the conventions used on the data sheets).
As to box manufacturers playing marketing games by lifting words they don't understand from the datasheets, nothing new there, but the chip datasheets are entirely normal and as ever are aimed squarely at engineers who read the real performance data not the stuff at the top of the page.
Regards, Dan.
Consider that the analogue headroom often exceeds that of the rest of the system, and a properly done digital gain control will track better then even an expensive pot...
Now consider that the conversion between a 32 bit value and a 24 bit one is not actually entirely trivial to get right (particularly for a floating point input) and actually being able to throw 32 bit data at the converter and let it handle the final gain control and conversion is actually something I would buy into if I was designing a processing box provided the analogue specs were competitive.
Converters have an ENOB (or noise floor, or noise power density or equivalent) spec for a reason and they all specify the input word length as the headline number (And have done forever).
Consider that outside the audio world, you can indeed get more then 24 bits of precision, it just requires a measurement bandwidth of a few tens of Hz, and that this is sometimes useful (And it is outside the audio world that defined the conventions used on the data sheets).
As to box manufacturers playing marketing games by lifting words they don't understand from the datasheets, nothing new there, but the chip datasheets are entirely normal and as ever are aimed squarely at engineers who read the real performance data not the stuff at the top of the page.
Regards, Dan.
Exactly, that's why digital attenuation should be avoided as possible, while otherwise loosing on the analog headroom.
32 bit full range
+----------------------------------------------------+
22/21 bit's = analog noise level
+------------------------------------+
Resulting maximal analog headroom
<------------------------------------>
Attenuated digital signal
........+--------------------------------------------+
Resulting attenuated analog headroom
........<---------------------------->
Hp
Here you are assuming that the final noise floor of the product is defined by the digital circuit or the DAC.
In the real world, if you put an analog volume control behind the DAC, you would not get any better results as the noise floor of the DAC itself falls well below the final noise floor.
What one should consider however, is matching the gain of the output buffer to the amplifier gain and the speaker drivers so you bring the noise floor as low as possible to begin with.
Indeed, the DAC is not usually the limit on system dynamic range (The room noise often is).
I still don't see anything inherently wrong with supplying a 32 bit input part with 21 bits ENOB in DC - 20KHz, and some on board signal processing, datasheets being doing that for years, and the consumers of the datasheets know exactly how to read that information.
32 bit interface, with some on chip maths = I can use a slightly cheaper DSP because I can offload some of the doings onto the DAC chip, maybe I can do away with the DSP completely!
21 bits effective = noise in DC - 20K is 123dB down = quite good enough.
That the end user marketing slime then take that data and misinterpret it for consumers should be no surprise (And is not the sand vendors problem).
Regards, Dan.
I still don't see anything inherently wrong with supplying a 32 bit input part with 21 bits ENOB in DC - 20KHz, and some on board signal processing, datasheets being doing that for years, and the consumers of the datasheets know exactly how to read that information.
32 bit interface, with some on chip maths = I can use a slightly cheaper DSP because I can offload some of the doings onto the DAC chip, maybe I can do away with the DSP completely!
21 bits effective = noise in DC - 20K is 123dB down = quite good enough.
That the end user marketing slime then take that data and misinterpret it for consumers should be no surprise (And is not the sand vendors problem).
Regards, Dan.
Very interesting discussion here about whether the manufactures lie about their DACs. The problem is, that there is no need for even 24Bit DACs. All you need is 16 Bit 44,1 kHz (Red Book audio)
The big lie begins, where sellers want to offer you High-Res audio files and 24 Bit DACs.
Just have a look at these links:
Xiph.Org Video Presentations: Digital Show & Tell
24/192 Music Downloads are Very Silly Indeed
Regards
Ralf
http://xiph.org/~xiphmont/demo/neil-young.html
The big lie begins, where sellers want to offer you High-Res audio files and 24 Bit DACs.
Just have a look at these links:
Xiph.Org Video Presentations: Digital Show & Tell
24/192 Music Downloads are Very Silly Indeed
Regards
Ralf
http://xiph.org/~xiphmont/demo/neil-young.html
My definition of a N-bit DAC is "N bits of the input value influence the output" - if all 32 bits are used from the input, with none thrown away or otherwise lost by the processing that occurs in the DAC, and you can demonstrate that changing the LSB produces a change in the data stream which feeds the output stage of the DAC, I'd call it a 32 bit DAC.
Though changing the LSB of the 32 bit input might only cause some little current source in the DAC output stage to hold a different value for <1uS once a week
Though changing the LSB of the 32 bit input might only cause some little current source in the DAC output stage to hold a different value for <1uS once a week
I dare to disagree even with Einstein, so why not with that site author...?
The problem with Science and understanding in general is, people always forget, or has INability to "remember" where assumptions have been made (in a theory) and how it affects reality IF the assumptions were wrong.
It is this very ability that will allow us to synchronize theory with reality. This ability will often appear as some one being open minded, knowing that there is room for "deviation" (can be measured in probability).
If people can imagine time travel or going through a wormhole, why cannot they be open minded to the possibility that more than 16 bit is critical? Don't they imagine what happen to human body when accelerating close to speed of light? In reality, so-called time dilation is so far from one second.
Few days ago I re-read about the first principles of Quantum Mechanics. So, in this "bullet" experiment it was found that an electron can behave like a particle, but it can also behave like a wave, but it is NOT either. There is a mystery about its behaviour. This is an example of ASSUMPTIONS I was talking about. Keep this assumption in mind and never be lost in sea of theory that is disconnected from reality. (So what is the relation with 16 vs 24 bit?)
I remember when I was young, I use my head as my phone book etc. "What is your phone number? Okay thanks", "What is the product key? Okay thanks". I remember when I was a kid and saw a rubics for the first time. It was so easy, I could imagine in my head how to solve it and I didn't loose track when I transformed the rubrics based on what I saw in my head.
When I'm old I was so sad that I lost my abilities. But there is compensation to the decrease in our physical body (neurons etc). We got KNOWLEDGE and EXPERIENCE. The smarter we are, the stronger the domino effect of this.
As to the ears, there is decrease in HF response but I know that I can "SENSE" it. I'm open minded that human might be able to "sense" very high frequency. Human might even be able to communicate with "non-human" using frequency above and below the "audible frequency".
Sampling theory is not rocket science. What critical is to know where assumptions have been made to allow for possible deviations. To see where theory might not in sync with reality. The reality is the "phenomenon". True, very easy to spot, many people never experience any "phenomenon". For example they cannot ABX 12/12 using Foobar, two signals that differs only in their absolute phase. Phenomena is required as a "check" for the accuracy of a "theory".
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.