FIR Digital Filter

[tmblack]

Its very easy to compute a FIR LPF up to half the sampling frequency.
Then convolve the filter coeff. with NOS signal.

Tom

EE

 

[rfbrw]

Back to PatPet. So you want to build a standard oversampling digital filter with the exception that the summing of the taps is done externally.
As an example I'll use the first stage of the SM5847. The first stage is a 2x filter. It has 169 taps. You could fold the filter but that would be cheating, so 338 dacs it is. And then there is the physical implementation of the filter. Still think this is a viable idea ?

 

[PatPet]

Back to PatPet. So you want to build a standard oversampling digital filter with the exception that the summing of the taps is done externally.
As an example I'll use the first stage of the SM5847. The first stage is a 2x filter. It has 169 taps. You could fold the filter but that would be cheating, so 338 dacs it is. And then there is the physical implementation of the filter. Still think this is a viable idea ?

For this number of taps, I barely think the impulse response will look good.

But I tried some FIR filter designers and found anything more than 3 taps will start resulting in pre-echo and post echo, which I think is no good at all. And for 3 taps we get 4x OS.

Maybe after that I play with linear interpolation and get more staircases on a 4X waveform. And hopefully the ultrasonics decay much faster.

And with 3 taps, the digital filter wont have good specs. But I'll trade something for good impulse response.

 

[voodooless]

Its very easy to compute a FIR LPF up to half the sampling frequency.
Then convolve the filter coeff. with NOS signal.

Sure it's easy, but what is the point in doing this? It won't even be a lowpass, since you don't have any audio data after 1/2 x fs anyway

But I tried some FIR filter designers and found anything more than 3 taps will start resulting in pre-echo and post echo, which I think is no good at all. And for 3 taps we get 4x OS.

Who sais it's audible? And in what freuquency range? With three tabs, I don't think you'll have an acceptable passband ripple. No way that's going to sound any good. You'll also need a speed enough stopband of course. The ringing will occur arround the crossover freqeuncy, and therefore with oversampling, you'll not hear any of it.

Maybe after that I play with linear interpolation and get more staircases on a 4X waveform. And hopefully the ultrasonics decay much faster.

Effectively, the FIR filter will do the non linear interpolation for you

And with 3 taps, the digital filter wont have good specs. But I'll trade something for good impulse response.

Just forget it, it's no good.

 

[PatPet]

quote:
Originally posted by PatPet
But I tried some FIR filter designers and found anything more than 3 taps will start resulting in pre-echo and post echo, which I think is no good at all. And for 3 taps we get 4x OS.

Who sais it's audible? And in what freuquency range? With three tabs, I don't think you'll have an acceptable passband ripple. No way that's going to sound any good. You'll also need a speed enough stopband of course. The ringing will occur arround the crossover freqeuncy, and therefore with oversampling, you'll not hear any of it.

Impulse Response

I think one of the offenders in OS is the pre-echo in the time domain and the other is jitter sensitivity.

I start digging myself into IIRs. Post echo is less offensive.

 

[voodooless]

I think one of the offenders in OS is the pre-echo in the time domain and the other is jitter sensitivity.

That's why the crossover is at such a high freqency where you don't hear it anyway. If you think it is audible, point me to a single decent oversamping DAC that has this problem. Good luck!

I start digging myself into IIRs. Post echo is less offensive.

Yes, but they are not phase linear, specially if you want a brick wall filter. Besides, there is still no use of such a filter if you don't oversample!

So explain exactly what do you intend to do with such a filter?

 

[PatPet]

That's why the crossover is at such a high freqency where you don't hear it anyway. If you think it is audible, point me to a single decent oversamping DAC that has this problem. Good luck!



Yes, but they are not phase linear, specially if you want a brick wall filter. Besides, there is still no use of such a filter if you don't oversample!

So explain exactly what do you intend to do with such a filter?

Okay I will answer this later. For the time being I find this quite interesting

http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf

 

[voodooless]

http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf

Upsampling != oversampling, the fact that the article does not seem to know the difference is not really good.

I will never claim that upsampling would have any positive effect on sound, oversampling just makes stuff a bit easyer. But the filtering just won't work with a non oversampling DAC, not FIR, not IIR.

 

[tmblack]

Sure it's easy, but what is the point in doing this? It won't even be a lowpass, since you don't have any audio data after 1/2 x fs anyway

No you have multifple images up to fs and beyond.

Maybe you should take a course or 2 in DSP.

Tom

EE

 

[tubee]

one could just build a filter using many DACs being fed digital signals with appropriate delays applied to the inputs.

I am that "one" at this moment, with 4 dacs. But Wadia and an other UK manufacturer used it also a long time ago.

And select difering sized summing resistors to create the tap weighting

Wa2ise, what do you exactly mean with this?

 

[rfbrw]

The Asylum. http://www.audioasylum.com/audio/digital/bbs.html

 

[voodooless]

No you have multifple images up to fs and beyond.

Sure, but the filter on non oversampled data will do nothing to get rid of that! Besides, you can hardly call that audio data, it's more like a ghost of it

Maybe you should take a course or 2 in DSP.

Maybee you should do the same. Are we now all going to any others with useless comments, or what? Please, just start a decent discussion, this kind of stuff is pointless and childish

 

[rfbrw]

So what data do you send to an oversampling filter ? I, in following in the footsteps of those who know considerably more than me, have always found it best to send non oversampled data to an oversampling filter.

 

[voodooless]

So what data do you send to an oversampling filter ? I, in following in the footsteps of those who know considerably more than me, have always found it best to send non oversampled data to an oversampling filter. [/B]

Of cource! But there is no point in filtering this data before you put in into the oversampling gear. That's the whole point I wanted to make...

I'm getting the slight feeling we all agree somehow, but are not making any sence to each other

 

[rfbrw]

Of cource! But there is no point in filtering this data before you put in into the oversampling gear.

But that is not what he wants to do. I think we are going to have to agree to disagree on what PatPet wants. He wants a oversampling filter but with the summing done in the analogue domain.

 

[PatPet]

But that is not what he wants to do. I think we are going to have to agree to disagree on what PatPet wants. He wants a oversampling filter but with the summing done in the analogue domain.

Actually I'm lost. With non-OS single DAC, we have slight roll off in HF. I can live with that. But the ultrasonics is what I want to remove as much as possible.

But then a good FIR filter which can attenuate greatly what I dont want, will lead to time smearing which people can argue it does offend or not. For me, I think later I'll test and trust my ears.

Then I start looking at FIR filter with slow roll off. That is a filter with small number of taps. And I start realising I will be doing something close to linear interpolation. To have linear interpolation, DACs are shifted by certain period of a sample. This has already been experimented by some members on this forum. The HF roll off is greater, although the roll off of ultrasonics is even greater.

 

[voodooless]

But that is not what he wants to do. I think we are going to have to agree to disagree on what PatPet wants. He wants a oversampling filter but with the summing done in the analogue domain.

Ah, I think I can agree on that, too

Then I start looking at FIR filter with slow roll off. That is a filter with small number of taps. And I start realising I will be doing something close to linear interpolation. To have linear interpolation, DACs are shifted by certain period of a sample. This has already been experimented by some members on this forum. The HF roll off is even greater, although the roll off of ultrasonics is even greater. [/B]

I'm not sure, but I can imagine that this method will also have some time smearing, since it comes very close the the fir filter priciples (it might even be exactly the same thing i a sense ).

Oh, well, I guess, I just don't get this NOS deal... keep it (presumably) simple... and then, make it a hell more complex anyway

 

[tmblack]

I don't know what are the intended fs or low pass freq and other criteria.

But I can assure you that we can make a 1kHz LPF with sampling freq of 8Khz non-oversampled. We can always take a few more DSP courses but I have taken enough in the past to have done this in the lab.

Tom

EE

 

[voodooless]

But I can assure you that we can make a 1kHz LPF with sampling freq of 8Khz non-oversampled. We can always take a few more DSP courses but I have taken enough in the past to have done this in the lab.

Sure you can do that!

 

[Petter]

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.

There are ways to design the filter with constant coefficients of say +1, -1 so this should be perfectly possible to do. I don't think it would be worth it though.