Optimum settings for recording LPs with further editing (sample rate and bit depth)

[Marc Hildebrant]

Craig,

I'm curious about the "brick wall" values. I would think that some type of filter is needed ahead of the A/D to prevent alias, but, I didn't think that the "passive" filters had different values. That would need some type of switching ?

Marc

[Craig Maier]

Marc,

I think that the A/D brick wall filter is usually done with a combination of hardware and software. The bricked wall values are usually just a tad below half the sampling frequency (to avoid aliasing), but in that one case (192 kHz) it was way off at roughly 21 kHz. 21 kHz only makes sense as the brick wall anti-aliasing filter for sampling rate values of 44.1 kHz.

I think that the sound card driver has a bug or something. It should have blocked my ability to record at 192 since it does not know how to handle that sample rate.

?

Craig

[Marc Hildebrant]

O.K., here is one experiment :

#1

Hardware: Sound Blaster Audigy with Marantz model 3800 and Grado Cartridge.

Record : Twelfth Street Rag, Dukes of Dixeland

Software : DC8

Setup : I used 44.1 KBits/sec 16 Bits and 96 KB/sec on the same song with the same multifilter. The multifilter uses the expert impulse and the EZ Impulse in a chain. The expert is 1/13/50 with the solo/brass on. The Ez impulse is 38.8/50.6/65.2 with the solo/brass on.

Results : The 44.1 setting causes trumpet distortion while the 96 does NOT.

This song has strong trumpet music and is a real bear to clean up. Yet , the 96 setting works "good" while the 44.1 clips and distorts the trumpet sound.

Marc

[Marc Hildebrant]

One more thing......


The 96 K settings left a big click in the song after running the multifilter, while the 44.1 K setting found it and removed it. So, it looks like the trade off is better sound with more clicks to remove manually...

Marc

[Craig Maier]

Hi Marc,

Great test. I wonder if that extra big click would be removed if the big click feature is turned on or the big click filter is added to the chain.

Craig

[Craig Maier]

Hi again,

Just been thinking about it - I wonder what 48 kHz sampling rates do in the context of your other two tests. I wonder if that value balances between better sound and click removal number.

What do you think? I would guess that 48 kHz is probably pretty similar to 44.1 kHz because of their closeness, but a test would bear that out one way or the other.

Craig

ps - this has become quite the long thread, maybe the longest here on this forum? Anyway, it is an interesting topic -

[Dan McDonald]

Craig -
When you did the test on your soundcard, did you use make waves or something to test for the ceiling or brick wall effects?

[Craig Maier]

Dan,

No - I just recorded silence with the input cable disconnected from the soundcard. The soundcard analog input circuitry has enough random noise to excite a frequency response graph. When looking at those recordings, the "brick walls" were real obvious for me to see.

Craig

[Craig Maier]

Marc,

When you ran your test, what mode was the Expert Impulse Noise filter set to? I guess it reasonable to assume Vinyl LP, but just wanted to be sure about that.

Craig

[Marc Hildebrant]

More Experiments......

#2 I tried using a sample rate of 48 KHz. Less trumpet distortion, but still some. Good click removal

#3

Using a audio generator, I fed in a signal directly to the line input. I kept the signal level constant (0.2V RMS) as I changed the frequency of the test signal and the sample rate of the audio module (sound blaster audigy).

I used the DC8 level indicator with the record function on pause.

1. 44.1 KHz falls off bad at 22 KHz input

2. 48 KHz falls off bad at 25 KHz

3. 96 KHz falls off bad at 22 KHz

****So...why is 96 KHz Sample better ?*****

Is it because there are more bits for the impulse filter to use ?

Marc

[Craig Maier]

Marc -

Wow - that is unexpected. I do know that part of the algorithm windows noise pulses based on the number of samples found. Maybe that is part of the explanation, meaning that a given number of samples at 44.1 kHz is "x" amount of time per sample (~22.7 uSec), but it is roughly "x/2" number of samples at 96 kHz (actually, x / (96 / 44.1). Otherwise, I do not understand the bandwidth aspect of things. Rick did go back and look at different parts of the algorithm, and my recollection about it was correct - - - higher sample rates provide better discrimination at ultrasonic frequencies.

I wonder why your soundcard is falling off like that. Can you try the motherboard soundcard and see what its frequency response is at 96 kHz sample rates? I just did that here on this machine (switched from the good soundcard over to the motherboard soundcard and looked at the response) and it surprised me. It went out to 48 kHz response (roughly) at 96 kHz. That is pretty good for a motherboard chipset, I think.

Craig

[Marc Hildebrant]

Craig,

I too was surprised at the Sound card's performance. One brick wall filter for all.

This machine is a real oldie, and the only sound card is the one in the PCI bus....

Marc

[Craig Maier]

Marc,

So, there is no motherboard soundcard (chipset) that you can test? hmmm I just pointed Rick to your post; I think that he will find it to be interesting.

Craig

[Marc Hildebrant]

Craig,

Yes, the system that I have been using is, shall we say, old. The system is a "power spec" system, circa 1998.
New CD burner,power supply,memory,sound card,usb,network card,etc.

While a new system is in the very near future, this one is it for now.

I think that the test data shows that the sound card has an anti-alias filter before the A/D that is fixed in value. The 96 K setting seems to provide more samples, but no real upper frequency gain.

I'm sure that the current crop of sound cards are much better !

Marc

[Craig Maier]

Hi Marc,

Yes, this computer is roughly 4 or 5 years old. Suprisingly, both soundcards here do actually have useful bandwidth to around 48 kHz with 96 kHz sample rates. But, we have not yet solved the riddle, so to speak - - - meaning - - - does ultrasonic response improve the discrimination capability of the noise reduction algorithm. The theory says yes (it is in the math within the routine), and yet you get better results without a dramatic increase in bandwidth (if any at all). So, there must be more to it than just the ultrasonic bandwidth - - - something with the number of samples per unit of time without regard for bandwidth (or maybe it is a combination of increased bandwidth and shorter timed samples).

I wonder if it is actually a combination wherein the absolute best performance really does occur when the samples per unit time are greater and the bandwidth is also greater.

Interesting.

Craig