workbench fun

[Tunnelman]

Eric

Working with the workbench is pretty cool, now i am starting to realize the concept and possibilities, great idea. I put the gain on the input side, and it seemed to help with the clipping. Is there any advantage you can see to this?

[eric_admin]

Tunnelman,

I'm glad you're having fun!

Yes, the workbench architecture is designed to be extremely adaptable. You'll find that you can do even do weird stuff like wire the knobs directly to the LEDs if you want, just to get a feel for how the system works. The knobs will trigger the on/off threshold of the LED input when they reach 50%. When the day comes that people write effects so complex they won't fit in a single cog, they'll find that the socket/conduit architecture can also be used to pass data between effects, or to pass pointers to shared RAM.

You can certainly put the gain stage on the input side of an effect if you like, but if you find that your guitar is consistently running "hot" and needs a gain reduction I'd be inclined to modify the OS to drop your gain at the input; that way you can remain compatible with the standard patch sets.

If you locate the two lines which read

Code: Select all

'Left justify incoming 24 bit sample
shl     in_data_right,#8


and

Code: Select all

'Left justify incoming 24 bit sample
shl     in_data_left,#8


in COYOTE1_OS.spin and change the "8" to a "7" in both places, you'll drop the input levels by 6db.

[Tunnelman]

Eric

Thanks for the info on the input level adjustment, i was trying to find that in the OS code, but there is lots of it, and i am kinda iffy on understanding the code anyway, but learning. Yeah, i have some custom hot wound pickups, and i am pretty heavy with the pick, so i will need to decrease the input for sure.
Really gets the wheels turning looking at the potential here.

Kel

[Tunnelman]

Eric

Sorry to be a PITA, i reloaded the OS with the changes and i am getting some major noise, almost like a huge overdrive happening. When the effect starts at the programmed gain it is normal, once i try to drop the gain the problem starts up. On a side note, if i drop the gain to below 10, it sounds like an amazing compressor, but still has some noise during silence. I reloaded it without the code change and it is fine.

[eric_admin]

Tunnelman,

No worries at all!

The Cause
Here's the full scoop. First of all, I have to apologize for not testing the fix I suggested more robustly. The fix has an error, which is that the 24 sample bits clocked in from the A to D chip are signed data, not unsigned data. If you change the left shift to 7 bits instead of 8 (per the fix), that leaves the uppermost bit a zero which is fine for positive values but it makes negative values positive which messes up the sound considerably EXCEPT when the final gain is set to 100% (which is where I had it when testing; that's default initialization in most patches). Why does it sound fine at 100%? The gain system resource is actually a cut/boost stage (It really should have been configured to display 0% to 200% because that is what it really does, which is where all this clipping trouble probably started; more on that later below). Setting the gain at 100% actually multiplies the output signal by two, which is equivalent to a single bit shift to the left, which gets rid of the offending incorrect bit (assuming you're testing with an effect which causes no net gain change, which I was), so you only hear the nasty distortion when you reduce the gain below "100%".

Two different ways to clip
There are two different ways to clip. One is to exceed the range of the input A to D which is 2V peak to peak. The second is to have have a net gain in the effect path which brings the output signal out of range.

If your guitar output runs so hot that it is clipping the input stage then no amount of post A/D processing will ever get the missing data back. The guitar needs to be turned down enough to avoid clipping the input.

Now, let's say you write an effect which has a gain of 400%, and your guitar puts out 2V peak to peak (full range). That effect will output a significantly clipped signal and no post-effect processing will restore the missing data. The solution is to write the effect to perform a gain reduction before processing.

Now, lets say you have an effect which has a gain of about 100% (most of the currently released effects), and your guitar is running full range, and you set the "Gain" system resource (the right most knob in all currently released patches) to "100%". That last gain stage actually runs 0 to 200% even though it displays as 0 to 100% (That was really a mistake on my part; I should have shown the range differently). So at a setting of 100% the gain stage is actually multiplying by two, which will cause the output to clip!. This is what is happening to you.

Two solutions you can use today
Here are two solutions you can use today:

1) Set the output gain knob to "50%". This will really give you a mathematical gain of "100%" (i.e. out=in), which generally will not clip the output as long as your input remains in range to begin with. The downside of this change is that you'd probably want to make it to all patches so that each patch loads with a 50% setting.

2) Modify the system gain stage so that it really does run "0 - 100%" (as displayed) instead of "0 - 200%" (as currently functioning). To do this, just comment out line 103 ("shl y,#1") of COYOTE1_Conduit_Engine.spin.

Code: Select all

                        'Signed multiply y<32> = (x<16> * y<16>)
                        test    x, sign_bit  wz
                 if_nz  neg     x, x
                        shr     x, #16
                        call    #mult
                        'shl     y, #1
                 if_nz  neg     y, y


The upside is that all current patches will now operate with a default output gain stage of 100% (which will match the displayed value when adjusting the knob). The downside is that you loose the capability to boost the output signal an extra 2x (6db) if needed (not such a big deal, really, since you can always turn up at the amp. It is handy sometimes to have the extra gain range when practicing through headphones.).

The long term solution
In the next Workbench & OS release I'll either change the output gain to truly run 0-100% (as indicated on the display), or to run 0-200% with a 100% default in all patches and a proper display range of 0-200%.

What about dropping the gain at the input?
Yes, it is still possible to drop your gain by 2x at the input processing stage, but that's generally not the best place to do it from a SNR (signal-to-noise ratio) standpoint. The fix above is better. If you still wanted to drop the gain at the input processing stage, you'd just have to add the line "sar in_data_XXXXXX,#1 " after each of the two existing "shl in_data_XXXXX,#8 " lines like this:

Code: Select all

                        'Left justify incoming 24 bit sample
                        shl     in_data_right,#8
                        sar     in_data_right,#1   


and

Code: Select all


                        'Left justify incoming 24 bit sample
                        shl     in_data_left,#8
                        sar     in_data_left,#1


The SAR instruction performs an "Arithmatic" (i.e. signed) bit shift, which will ensure that the upper bit is maintained properly for the sign of the value.

[Tunnelman]

Eric

I started to suspect the condition of the wiring in my guitar, after all it is 35 years young now, and it has gone to the shop for a makeover. I got a loaner strat for the meantime, and i don't see any clipping with the original code. The pickups on the strat are lower output than humbuckers in general, so that will make a difference anyway. It will be a few weeks until i get mine back to try again. Anyway, nice sound with the strat.

[Tunnelman]

Update _____ Yep, it was the old condition of the wiring / pickups that was causing the issue, all is good. My apologies for the runaround.