Recently I read quite a number of articles on digital filters, especially FIR ones. In a FIR filter, there are taps in which digital signals are delayed for certain period of time before they come out multiplied by a (unique?) coefficient and are summed as a single output. The result is there are new interpolated samples which smooth out the staircases and thus the analog filters can be of lower orders.
This is how a common digital filter works. Usually the filter chips has limited internal accumulator length which may be a problem. And usually there are stages of taps. For example, 2X filter followed by 2X filter followed by 2X filter. This according to Ryohei Kusunoki will cause audible accumulated delay.
Is it possible to design a single stage FIR digital filter that doesnt sum the output, and in other words the individual lines from the coefficient mulitipliers are fed to individual DAC chips?
This is how a common digital filter works. Usually the filter chips has limited internal accumulator length which may be a problem. And usually there are stages of taps. For example, 2X filter followed by 2X filter followed by 2X filter. This according to Ryohei Kusunoki will cause audible accumulated delay.
Is it possible to design a single stage FIR digital filter that doesnt sum the output, and in other words the individual lines from the coefficient mulitipliers are fed to individual DAC chips?
If both channels of the audio are filtered the same way, that won't matter. Sure, there's delay, but a few fractions of a second are of no consequence. As for truncation effects, a good filter design maintains all the bits, and rounds things back to 16 bits (or more if the DAC chip can accept them) just before they hit the DAC chip.
If the coefficients are carefully selected, one can build filters without having to use real multipliers. Simple combinations of powers of 2 are easily done with shifts and adds.
PatPet said:
Why would 1x fs have less jitter than a Xx fs DAC? Besides, it will not work! You can't filter digitally wil a 1x fs DAC, since the output of the filter is still 1x fs, the filter is of no use.
You need Xx fs to gain bandwidth, and then filter it down with the FIR filter.
Not more processing power, but more memory. Also, this is were oversampling comes in handy, since at higher frequency, you'll need less tabs for the same result as a comparable filter at low frequency.
The output of an ordinary digital filter is the summed signal from all taps.
What I'm interested is if we dont sum the taps and feed the DACs with individual signals from the taps and sum finally at the outputs of the DACs, we'll get an analog signal that looks like it has been oversampled. If a DAC is fed with 1Fs signal, it is less sensitive than it will be if fed with oversampled signal.
rfbrw said:
Well, that's to bad then!
Let me explain why (I think) I am right:
What is the digital filter for: wel, exactly the same thing as the analog one would do, but this one works in the digital domain. So: we need to filter somewhere on the end of the digital bandwidth of the orriginal signal, lets say 20 Khz. Now we apply our 20 Khz FIR filter to the non oversampled signal and get a 20 Khz filterd non oversampled filter again. So now: what use is this? Answer: non what so ever! For the filter to work you would need a stopband bandwidth after the filter, and with non oversampled data, you don't have this.
If you thinkt is stupid or wrong, please elaborate, because with comments like this, you can better stay away for this forum completely.
Exactly! So it does exactly what a normal analog filter would do, only in the digital domain. Obviously you'll still need an analog filter, but it can be at a far higher frequency.
So what! Since when does that mean anything? If you think I'm wrong, come with decent arguments. We're all here to learn, so and if I'm wrong, I want to be told, not called a stupid wiseacres (whatever that means...)!
It's to bad you didn't comment on the rest of my post, since that is the most important part!
rfbrw said:
No it's not! The topic starter want to do FIR filtering in a non oversamping DAC. I say: you cannot do that for the reasons I already stated. Sure you can put an external ovesampling fiter in front of your (whatever) DAC, but that will do 4 to 8x at max. Imho, that is not really a lot, and I don't think It's of any use.
rfbrw said:
You still don't get to, do you! He want's to use a nos DAC, and want to use the FIR filtering (combining the (somtimes) present three stage filter).
So actually, it IS about nos DAC's, and this would mean that FIR filtering is of totally no use. That you can combine the three stage filter (if present) to just one is of course obvious, and is already said.
The actual problem is that the topic starter doesn't completely get that the FIR filter tabs are. They are no seperate signals, but just coefficients that are used with multiplication and delay (just to put it very simple just now), to create a filter.
The higher the frequency, the less tabs you need, meaning less delay. And also the higher sampling will help here, since it means you can process more tabs in the same time, resulting in less passband ripple. So it actually is a balace between speed, delay and quality.
The three filter in series will have not more or less delay than a single filter with about the same resulting response. It will however takes mess memory, since you can use three smaller tabs, and not one larger one I guess. But I also guess, that not all DAC's will use a three stage filter.
About the delay beeing audible: it's not of course, since there is no phase variation in the filter the delay is exactly the same at every frequency, and for both channels. There is no way you'll hear it. It's as if you would play the CD a few ms later that you inteded to do.
@Nemophyle: don't argue with rfbrw. You have been arround for even a shorter time than I did. you couldn't possibly know what you are talking about
If you don't use a tap's output, you effectively set that tap to zero.
Yes, one could just build a filter using many DACs being fed digital signals with appropriate delays applied to the inputs. And select difering sized summing resistors to create the tap weighting. Nobody does that, it's much cheaper and more accurate to upsample the digital audio and digitally filter that and then feed it to a higher percision 18 or 20 bit DAC. You'd still need the 8x or such system clock anyway, to provide the delays in my first circuit above. You just select a higher frequency crystal oscillator and use flip-flops to divide it down (very low jitter, not like PLLs).
I've forgotten why oversampling is considered bad anyway...
Yes, one could just build a filter using many DACs being fed digital signals with appropriate delays applied to the inputs. And select difering sized summing resistors to create the tap weighting. Nobody does that, it's much cheaper and more accurate to upsample the digital audio and digitally filter that and then feed it to a higher percision 18 or 20 bit DAC. You'd still need the 8x or such system clock anyway, to provide the delays in my first circuit above. You just select a higher frequency crystal oscillator and use flip-flops to divide it down (very low jitter, not like PLLs).
I've forgotten why oversampling is considered bad anyway...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.