Meridian Unplugged - Powered by MrTechGuy.co.uk

Thanks for your comments Charles, I did'nt know you were (also!) around here... it is good to have such a big guy from the 'industry' in advanced forums like this one, diyaudio or audioasylum.

Hi Charles,

I'm getting to grips with the MDS program and have had a good play. The following parameters seem to approximate Ayre's MP "Listen" filter.

Lowpass Filter
Sampling freq: 88200
Passband freq: 1000
Stopband freq: 44099 (i.e. use entire audio band)
Passband ripple: 1dB (not quite sure what this does)
Stopband ripple: 144dB (ensures at least 144dB (24-bit) of ripple rejection?)

Interesting how low the passband frequency had to be to achieve the desired curve (which corresponds to approx. -1dB@16kHz and -3.2dB@20kHz). I guess it's only a parameter...

Can you comment on the issues involved with allowing aliasing in the entire audio band?

Edit: I just re-read this comment:
I'm now confused. How can the filter I generated not be aliasing?

Thanks,
Joel

It's quite a good program in my estimation. It takes a while to get used to it, but I think that is true of all computer programs.


Some of the terminology in MDS is straightforward and other is less obvious.

Sampling freq: This is obvious.
Passband freq: This is not so obvious. It must apply some criteria such as the frequency at which you get -0.001 dB of rolloff or something. But this in turn also depends on which windowing function you have choose in a later step.
Stopband freq: This is the frequency at which the filter reaches its maximum attenuation (specified below).
Passband ripple: This is the allowable ripple in the passband. I usually just set it to something arbitrarily low such as 0.01 dB or 0.001 dB. The final filter will normally have much lower passband ripple values than what I specify, so I don't pay a lot of attention to this one.
Stopband ripple: This is fairly misleading. It is actually the minimum stopband attenuation value.


The program is aimed at designing tradition "brickwall" filters. When you start to attempt slow rolloff designs, you have to fudge these factors, just as you are doing.


I think you are overstating the problem. First look at the attenuation of the filter. Even a slow rolloff type will have 10 - 15 dB of attenuation at 24 kHz, which is where the first image of a 20 kHz audio signal will appear with Redbook format. But then you also have to ask what is the spectral content of the music signal at 20 kHz. Say it is down 20 - 30 dB from full scale. Then the aliases would be down by between 30 and 45 dB, at a frequency of 24 kHz. As the frequency goes up, the level of the aliases goes down.

That's why it's not such a big problem in the real world, and why Wadia successfully built their company on that approach.

Finally, it's all a trade off. Remember I told you to keep the impulse response graph open. The impulse response of a slow rolloff filter is literally 10x better than a standard "brickwall" filter. With Redbook there is no "right" answer because the format is flawed. But only about 2% of our customers prefer the brickwall filter to the slow rolloff filter.


The filter will allow aliasing. Virtually every filter ever used in every CD player ever made has aliasing. They were all "half band" which has half the taps and half the cost, but is only down -6 dB at Fs/2. With only one exception (a Cirrus Logic A/D chip) the so-called "apodizing" filters are the first ones that try to eliminate aliasing. Even these can't totally eliminate it. They may reduce it 80 or 100 or 180 dB, but one can never completely eliminate it.

But the program won't let you select a passband frequency higher than Fs/2. That's what I was referring to before. Sorry for the confusion.

Thank you so much for this contribution Charles. As I'm not in the industry and there is probably no definitive book on the subject, it's often really difficult to pick up know-how gained via years of experience. I've been scratching around Wikipedia, white papers and the like, but your input has really helped. This is all making so much more sense.

I'm working on the sox guys to allow arbitary rolloff in the filter...

You're doing exactly what the best guys in the industry do -- try a bunch of different ideas and see what sounds best. I have found the MDS program to be a super useful tool that allow you to see exactly what tradeoffs you are making. You can look at a half-dozen filters in 15 minutes or less.

You can't overlay plots, but you can print them out. Then overlay two and hold them up on a window to see how they compare.

It's nice that Sox is small enough that the guys will listen to you! If they were a big company, you would have to be a big customer before they would even answer your e-mail.

Think about contacting Texas Instruments and requesting a feature in a DAC chip. If you're Sony or Panasonic, they'll listen. If you're Ayre or VirusKiller, they would just laugh.

Just to report that I didn't get a positive response on the SoX users DL when I asked for pointers to how I might modify the (undocumented) source code. So I'm going to have to wade through the source code myself, which is going to take some time...