I dont know why you cant see the contradiction. You said everything is analog in a ClassD amp (...) that accepts S/PDIF signal. Do you really think the digital signal magically transforms itself to analog as soon as enters to an amplifier? Because if not, then there is a digital signal inside, which is falsifying your statement.
D
Deleted member 148505
So "Everything in a switching output stage is not digital"?
Still something not defined: what is "volts, time, amps"? "Volt" alone is not a defined thing. For example: voltage of what? And the only relevant question here is: "time" Time itself is continuous. But can switched voltage change polarity any time really? In a clocked logical system: no. The timing of rise and fall is restricted to discrete steps. This makes it digital.
Still something not defined: what is "volts, time, amps"? "Volt" alone is not a defined thing. For example: voltage of what? And the only relevant question here is: "time" Time itself is continuous. But can switched voltage change polarity any time really? In a clocked logical system: no. The timing of rise and fall is restricted to discrete steps. This makes it digital.
Aha, you mean a system/box/circuit (the TI one) with a digital input and a class-D output stage.... OK. One day you will maybe get it.
A "power DAC" like Tri-path is a true digital power amp. Just because you combine a digital input, a D/A converter and a PWM output stage in one circuit dosent make PWM / Class D digital.
//
Switchmode resolution.
Here's my view on this, correct me if I'm wrong. In a multi-bit sampling system (16 bit for example), the clock rate will be much lower than in the 1 bit Delta-Sigma or PWM, so aliasing distortions might be the main challenge to deal with. Aliasing is where the output frequency folds back, relative to the input frequency, when there are less than two samples per period of waveshape. If the sample rate is 44kHZ, and you feed in 25kHZ at the input, you'll get 19kHZ out... Anything higher than half of the sampling frequency gets folded back. An analog reconstruction filter would need to have a VERY sharp cutoff rate, in order to allow decent audio bandwidth, and keep aliasing distortions low enough, with a 44kHZ sampling rate.
PWM and Delta-sigma one bit systems will be limited by rise time, rather than aliasing distortions since their sampling rate might be around 1mHZ, and is therefore more likely to have slewing related distortion issues. They sample fast, but can only charge the reconstruction cap so much per sample. Both of these issues were very challenging back in the 1980's, when processing had big limitations on speed and sophistication of processing in the digital domain (such as up-sampling). Any more, I'm surprised if these issues are significant.
In a switchmode (class D, etc.) power amp however, reconstruction of the switchmode output is done only by a 2 or so stage analog passive filter (coils and caps). This method has limitations that will compromise the fidelity some. Any D to A will have these issues to deal with, but a switchmode poweramp is also driving a reactive load, which can affect in realtime, the effect of the reconstruction filter (pole vibration). If the passive reconstruction filter is designed with more sections in order to achieve a higher cutoff rate (4th order for example), it will be even more sensitive to this reactive load impedance.
Whether this is anything to worry about I don't know, but it's what I'd look close at. I've read that some switchmode power amps can become unstable with either no load or too much load, which could blow up the speakers, so I'm still a bit hesitant to go after switchmode poweramps unless I think I can really trust the brand name.
In a portable unit, such as something you could take camping, I'd definitely go switchmode for the efficiency (battery life), and just not spend too much on the associated speakers.
Here's my view on this, correct me if I'm wrong. In a multi-bit sampling system (16 bit for example), the clock rate will be much lower than in the 1 bit Delta-Sigma or PWM, so aliasing distortions might be the main challenge to deal with. Aliasing is where the output frequency folds back, relative to the input frequency, when there are less than two samples per period of waveshape. If the sample rate is 44kHZ, and you feed in 25kHZ at the input, you'll get 19kHZ out... Anything higher than half of the sampling frequency gets folded back. An analog reconstruction filter would need to have a VERY sharp cutoff rate, in order to allow decent audio bandwidth, and keep aliasing distortions low enough, with a 44kHZ sampling rate.
PWM and Delta-sigma one bit systems will be limited by rise time, rather than aliasing distortions since their sampling rate might be around 1mHZ, and is therefore more likely to have slewing related distortion issues. They sample fast, but can only charge the reconstruction cap so much per sample. Both of these issues were very challenging back in the 1980's, when processing had big limitations on speed and sophistication of processing in the digital domain (such as up-sampling). Any more, I'm surprised if these issues are significant.
In a switchmode (class D, etc.) power amp however, reconstruction of the switchmode output is done only by a 2 or so stage analog passive filter (coils and caps). This method has limitations that will compromise the fidelity some. Any D to A will have these issues to deal with, but a switchmode poweramp is also driving a reactive load, which can affect in realtime, the effect of the reconstruction filter (pole vibration). If the passive reconstruction filter is designed with more sections in order to achieve a higher cutoff rate (4th order for example), it will be even more sensitive to this reactive load impedance.
Whether this is anything to worry about I don't know, but it's what I'd look close at. I've read that some switchmode power amps can become unstable with either no load or too much load, which could blow up the speakers, so I'm still a bit hesitant to go after switchmode poweramps unless I think I can really trust the brand name.
In a portable unit, such as something you could take camping, I'd definitely go switchmode for the efficiency (battery life), and just not spend too much on the associated speakers.
You're right I made a mistake between AD and DA. I should have re- read my post one more time. But this doesn't change the fact that both types of converters DA and AD are mixed-signal devices simply due to the fact that they do the conversion between both domains.
I would simply call a class-d amp a class-d amp or switchmode amp. One could of course use more detailed descriptions if the target audience is more technically oriented (natural sampling PWM, self-oscillaing PWM, delta-sigma, .......). But I would not use the marketing BS term "digital amplifier" not even for the the ones like the TI amps with their digital modulators, which use oversampling and noise-shaping together with low-resolution PWM in order to achieve high resolution in the audio band.
Regards
Charles
You are right, in this meaning a ClassD amp can be also called a mixed signal device. Unfortunately this terminology question is already answered differently by the developers of such "amplifier" (you can also say, it is not an amplifier, since the input and output is in different dimensions: digital numbers sv. voltage). But this is a completely different question from the previous debate: Is there any digital in it, or not.
And it's perfect. However telling TI call their amp incorrectly Digital, is a different thing and requires more reason.
It is you choice not to tell something. Others choose differently, and I see reason for it: it is quantized, and uses digital processes. Up to the last stage it is fully digital. Maybe you also have some logical reason for denying the term digital.
BTW: thinking "Digital" means "better than Analog" is simply ignorance IMHO.
I've already seen on the first page that you are wiser than the persons trying to teach you.
How else would you get an analog sound from a 16 bit, 44khz PCM stream that is the input to such a chip?
//
Last edited:
I've already seen on the first page that you are wiser than the persons trying to teach you.
And the blind lead the blind.
//
I understand. So if you know only one way, then it must be the only way you know. Learn digital signal processing and you will see!
There are materials about digital PWM modulators, for example: http://ww1.microchip.com/downloads/en/DeviceDoc/31014a.pdf
And please don't say "ahhaaa, there is a comparator in Figure 14-2, so it must be analog", because if you get to the Figure 14-3 you can see that comparator is digital, with 2 digital input word and a 1 bit output indicating the two 10 bit wide words are equal.
This is NOT what a typical Digital amp made of, only an example for you to show 1 kind of lego brick to build some kind of digital amp. There are more ways. (However at the end of the learning curve you might be able to see the principle is the same for all.)
---
I always think people can learn no matter how many counterexample I experience. I'm this kind of idiot.
As PCM (in) is digital and PWM (out) is analog, there need to be a mechanism to convert a number to a corresponding level (of something) - hence the need for a D/A. There is no way a PCM can be turned into a analog signal without D/A conversion I hope we can agree on that at least. You are confusing a 2 level signal, (sometimes called "binary" which is a number base and not something that can adopt only 2 states) with "digital". In a PWM amplifier there is no quantization as in sample and hold and therefore nothing digital about it and with this follows that the resolution is basically limited by the noise on it's output.
//
//
Last edited:
Bob,
There are many different questions discussed by you. Many are well known and analized already by Nyquist, Shannon, etc, and the metodology of checking/avoiding these artifacts is clear and "standardized". You missed mentioning only 1 thing: quantization, the operation responsible for loss of resolution.
Sampling. (This is not a question about resolution, so it would be better to discuss elsewhere.) To avoid aliasing input signal must not contain any component over the half of sampling freq. This is ensured in almost all cases not by very steep analogue filters, (because these ruin phase response or even many other parameters), but by a simple analog filter, oversampling, digital filtering, and decimation.
In case of ClassD a bandwidth limit is also required, typically 1/10th of fsw is allowed. Check my post #13! It is a requirement, but in most times it is fulfilled. And bandwidh limit together with signal amplitude limit also imply a slew rate limit. Is it possible to build a ClassD amplifier that can be driven into slew rate limit with a signal freq below transfer bandwidth? Yes. I also designed such amp. But then this must be specified, like "small signal bandwidth: 80 kHz, power bandwidth: 30 kHz at 100W at THD=1%". And this is not a ClassD specific behaviour. And no connection with resolution. It would be the opposit direction: low signals.
Reconstruction
Charging speed is not limited by pulse freq, but inductor and supply voltage. Actually it is an LC filter, not a cap.
Sidenote: unless input antialiasing filter, omittion of reconstruction filter doesn't cause any artifacts inside baseband. Some PWM amps not use any reconstruction filter, and only the HAM radio operator can tell it. Reconstruction is only aesthetical. It pleases only our eye, not your ear.
More to come, maybe.
There are many different questions discussed by you. Many are well known and analized already by Nyquist, Shannon, etc, and the metodology of checking/avoiding these artifacts is clear and "standardized". You missed mentioning only 1 thing: quantization, the operation responsible for loss of resolution.
Sampling. (This is not a question about resolution, so it would be better to discuss elsewhere.) To avoid aliasing input signal must not contain any component over the half of sampling freq. This is ensured in almost all cases not by very steep analogue filters, (because these ruin phase response or even many other parameters), but by a simple analog filter, oversampling, digital filtering, and decimation.
In case of ClassD a bandwidth limit is also required, typically 1/10th of fsw is allowed. Check my post #13! It is a requirement, but in most times it is fulfilled. And bandwidh limit together with signal amplitude limit also imply a slew rate limit. Is it possible to build a ClassD amplifier that can be driven into slew rate limit with a signal freq below transfer bandwidth? Yes. I also designed such amp. But then this must be specified, like "small signal bandwidth: 80 kHz, power bandwidth: 30 kHz at 100W at THD=1%". And this is not a ClassD specific behaviour. And no connection with resolution. It would be the opposit direction: low signals.
Reconstruction
Charging speed is not limited by pulse freq, but inductor and supply voltage. Actually it is an LC filter, not a cap.
Sidenote: unless input antialiasing filter, omittion of reconstruction filter doesn't cause any artifacts inside baseband. Some PWM amps not use any reconstruction filter, and only the HAM radio operator can tell it. Reconstruction is only aesthetical. It pleases only our eye, not your ear.
More to come, maybe.
One can hear down one Bel or so into noise floor AKA hiss. But a signal one Bel below sampling threshold of an A/D-converter is lost forever. This is, why dither helps.jan_didden said:
All amplifiers except for Class-A have a tendency to cut off signals below a certain threshold. This is, because power actors (actually MOSFETs, but vacuum valves and bipolar transistors are similar) tend to ignore small signals. Class-D is PW(M/DM) on top of Class-B, meaning that below its switching range it behaves like Class-B.
Switching range equals 2Pi comparator frequency divided by switching time of power actors. Switching range is around 4 to 8 Bel. Say the Class-D device produces 2 BelVolt at maximum, and its switching range is 6 Bel, then it becomes Class-B at levels below -4 BelVolt.
In order to lower Class-B distortions, tricks such as idle current, feedback and feedforward may be applied. Some modules have heatsinks and idle currents of around 100mA. Others have fast power actors and so great switching range. Yet others use feedforward.
Finally, most modules apply a dither to mask nonlinearity at low levels. Therefore Class-D is usually noisier than Class-A/B.
This is simply false. Both generally and in every details. No such rule. "Switching range" is also a nonexistant thing figured out by you and even according to your own idea it is faulty. This is only a play with numbers without any consideration based on ClassD topology and circuit analisation. Analization can easily show it is false. There is a small signal nonlinearity in ClassD, but at a completely different domain. It is called dead-time distortion, and there are 2 domains above and below zero output current where arises. It is where output current reaches the peak current of the inductor ripple. Between these 2 domains the transfer function is linear. This small signal domain is amost perfectly linear, and reacts to the smallest changes in control signal. The nonlinear domains are depending on inductor value, and typically are at 10% of max output current. In most cases THD plot shows this.
Except for wrong terminology of ClassB at least this is true. However idle curent is not a very informative term here. It can be either cross-conduction or reactive current.
You must have analysed many modules. Can you show only 2 in which dither is used intentionally and this info is publicly accessible?
Last edited:
I think the title of the topic must be: "People who never developed any ClassD and didn't learn about it either! Please discuss your self-made theories and definitions!"
And this is good. There must be a playground for them with a high noise floor even if they don't understand what it means.
And this is good. There must be a playground for them with a high noise floor even if they don't understand what it means.
I thought you meant before the PWM output stage.
From input to PWM stage there is no digital to analog conversion. But there is digital to digital conversion. I'm not 100% sure, but I think the DA conversion takes place after the PWM output stage in the reconstruction filter
http://www.eetasia.com/STATIC/PDF/200909/EEOL_2009SEP04_ACC_TA_01.pdf?SOURCES=DOWNLOAD
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.