Stereo Tool

lol.. i somehow managed new clipper to work at transmitter site with E5400 CPU, with declipper On and clipper strictness at 13.
Also, i think i heard at one moment some clicking.. (Hans told me something about click every 46ms) - wanted to record it, but after 20-30 seconds, gone.

I think I may be able to solve the clicking after all - I thought it wasn't possible but I seem to have been wrong about the exact cause (I already reduced it by a few dB tonight, just by changing a few values).

Problem:
- For clipping, I process a block of audio, say 4096 samples
- To separate this into multiple threads, I split this block into 2 2048 sample blocks. (They have to be a power of 2).
- Problem: When I merge the two together again, you would get issues at the center. So some (small) amount of overlap is needed.
- Solution: The start and end of the 4096 samples contain near-silence, so I thought it doesn't matter much what you do there. So I basically 'forget' a few samples at the start and end of the 4096 samples block to be able to have some overlap.

In the end, what I do to fix this is just clip the samples at the start and end using 'traditional' clipping, even without oversampling.

Now it turns out that if I make the overlap area smaller, the ticks are softer.

I also found that I did something wrong in the merging code at the center, when I fixed that about half of the clicks disappeared immediately.


Now, I could use the same clipping method that I use for normal processing also for the few samples at the edges. But the amount of power in it is so small that I really don't expect much from it. Also, any audible clicks (as Bojcha just reported) may have come from the wrong merging code.

What I can at least do, it's simple and should be reasonably effective, is to replace the current clipping method by something that takes the stereo and RDS signal into account. I'll add that and I think I'll leave it there unless people really hear clicks with the new version that I'll build tonight.

Lol. Ok, so that wasn't the cause at all.

Turned out that I was doing something wrong in the demodulation code: I didn't remove the pilot and RDS information from the signal before demodulating. And the RDS signal was interfering with the audio. For some reason, that apparently wasn't an issue in the singlecore version, but it was in the multicore.

There are still some differences in the audio between the singlecore and multicore versions, but no clicks anymore.

Did some more measurements. Stereo separation in SSB mode is indeed not very good (still above 30 dB, so I doubt anyone will notice, but it should be more - for DSB mode it's around 48 dB). I think I know why - right now I use a bandpass filter to remove the USB (Upper Side Band), and increase the LSB (Lower Side Band) to compensate. But if the two filters don't precisely match - and they don't - this leads to all kinds of issues.

I'll have to start using a Hilbert transform instead: Using the Hilbert transform I can directly generate the signal the correct way instead of fixing it afterwards.

Anyway, I would say SSB is usable for now, just not perfect.


I'll start a build now; ticks are gone in multicore mode.

Dual core clicking is gone now.

Stand alone: http://www.stereotool.com/download/ster ... 20-007.exe
Winamp DSP: http://www.stereotool.com/download/dsp_ ... 20-007.exe
VST: http://www.stereotool.com/download/vst_ ... 20-007.dll

IMPORTANT: To get a good sound from the composite clipper:
1. Enable it It's an option under Advanced Clipper
2. Under Advanced Clipper -> CPU, set 'Strictness' to 20 (at least above 10; check how much your CPU can handle) and Stronger Clipping to around 25-30%.

Warnings:
You will need a license update to use this! Otherwise the 'unregistered options in use' thingy will pop up, and you'll get messages through the music (every 4 hours).


The bug I reported earlier appears to be VMware/Windows API related, so I'm going to ignore it for now.

I may be able to reduce the CPU load a bit further (for both the composite and the normal clipper!).

'Clipping Strength' currently is a bit annoying: Set it too high and the audio may be damaged, set it too low and clipping is ineffective (increasing Strictness helps, but it increases the CPU load).

I think I can make it calculate the 'optimal' strength level on-the-fly, which should mean that Strictness can be lowered.

btw Hans - what about value strictness CPU and stronger clipping in standard clipper without stereo encoding and rds, set it to maximum if possible on my CPU or how to set it optimally for best sound?

_________________
Urban Junglist

Higher strictness = tighter clipping. But, in normal mode it settles down much faster (less Strictness needed) than in Composite mode.

The default values should be pretty much ok, to test turn of Hard Limit and look for spikes above the maximum level. Some small spikes will always be there (Hard Limit will remove them), but the combination of Strictness and Strength determines how big they are. For best sound, it's better to have a higher Strictness level with lower Strength than the opposite.

AMD X4 9950 = No Way!

Thx, now I know
I'd like to now how "take some shortcuts" and "take oversampling shortcast" affect sound
"take some.." default is on but description "reduces quality"?

_________________
Urban Junglist