Flat Filter & Unexpected Peaks


I realise that there’s some “unexpected boost” in the output signal while filtering sounds with the *Filter DSP, even if it has a flat curve and even without any resonnnant curve.

Check this .xrns example. There’s a BW 4n Filter with a 100% flat curve at the end of the Master.

Turn the Auto-Gain (AGC) ON ; then, start the track playback with the Volume to the Max. The first pattern (without the *Filter DSP) shows no Auto-Gain action.

While playing the second pattern (that is 100% identical to the first one, but that only has a *Filter DSP turned on), a sudden Auto-Gain action happens, compensating a strange peak of 0.812dB on the Master. :huh:

? what’s happening there ?

(could it be that there’s a kind of dry/wet internal problem? or something else?)

Not seeing anything. The slider remains on 0dB at all times.

I’ve just tested my example on all my 3 different computers at home (laptop, desktop, with/without directsound, with without ASIO, with/without more latency, soft clipping)… and when I start my track at 0dB, I clearly get peaks when the *filter DSP is turned on.

Did you turned the Auto-Gain On at the beginning ?
Did you see some kind of red peaks while playing the 2nd pattern ?
Do you run the example on Windows / Linux / Mac ? 32 bits ? 64 bits ?

that is so weird… I totally see the auto-gain compensating

You’re expecting a digital filter to be mathematically perfect, when in reality they’re usually quite far from it :)

(I also see the peaks you’re talking about)

Given totally ideal conditions, the Butterworth filter is designed to have a “maximally flat” response with no ripple. However, since computers are far from ideal and don’t have infinite frequency bandwidth or processing power, compromises must be made somewhere. These digital limitations will often result in slightly strange behaviour from filters, especially when you’re operating at the very extremes, ie. high-passing at 0Hz or low-passing at 22,050Hz. You’ll also notice that setting different sample rates in Renoise will produce slightly different gain variations/ripples in your test. Generally speaking, you should simply expect some amount of imperfection. But that’s ok - we’re making music here, not performing critical scientific signal analysis :)

Funky little tune btw! Nice use of my HAM radio sample :)

Thanx :)

In fact I use more & more the new *Exciter DSP, that really add something fresh in my last mixes. But since I also love the *LofiMat & the *Distorison DSP, my sound is sometimes “sharpened” so well that it becomes very irritating, especially in the highest frequencies bands. I had the idea to fix it, using *Filter DSP, cutting everything that’s too harsh after 20Khz / 21Khz, at the very end of my chains in the Master track ; but I realised that doing so I encounter those peaks problem, what leads me to lose a little bit of some “dynamics range”… Well, no problem if it’s something that can’t be worked out easily ; I just moved my *Filter BEFORE the *Maximizer, (and not AFTER), and everything works fine then, no peaks anymore.

Thanx dBlue !

Multiband Send + Distortion = Multiband distortion :yeah:


The AGC is turned on in your song, did not have to.

No red peaks, not compensation, nothing.

Was testing on Renoise 2.8 64bit on W7.

Thanx :) Renoise 64bits seems to produce more acurate *Filtering results than Renoise 32bits. Anyway I’ve found a valid workaround for my problems.

Not sure if 64 bits should make a difference here, I think it says somewhere in the docs that 64 bit version uses the same DSP engine as the 32 bits.