Synthesizing a 909 Kick

[saint_stink]

There seems to be a lot of wistful talk like “I hope Justin Valer releases a 909-style AR kit like his 808 one!” going around here. While I too hope for the day a true master takes on the 909 using purely AR synthesis, I figured no better time than right now to start a dialogue about synthesizing the 909 kick on the AR (no samples!).

Today I took a crack at it and want to report some results:

First, I’d like to refer everyone to the terrific research and analysis performed by Gordon Reid in the Sound on Sound article on bass drum synthesis

It provides a detailed synth architecture for the 909-style kick. However, I did not see any citation to an actual Roland schematic/design nor does he provide an audio sample, but perhaps this is all general knowledge about the 909.

As a reference to the actual sound, I used the original 909 kick included in Elektron’s Divine 909 sample pack (I really hope this is a decent reference because I’ve never owned a real 909 myself, but I personally really like this kick).

First, I attempted to emulate the kick using each of the AR bass drum machines and after wrestling with each of them, I was most satisfied with the results from the Hard BD, Classic BD, and Plastic BD.
However, each of these machines lack the low-passed white noise transient present in the 909. To remedy this, I used the noise machine on another voice through a 2-pole low-pass filter to augment the kick. If you have a MK2, I recommend sampling this noise transient and playing it through the kick’s sample engine so that this patch doesn’t require more than 1 voice.
While each of these patches are definitely 909-esque in nature, they still couldn’t match the crunchiness of my reference 909 kick. But perhaps this reference sample was heavily processed, it certainly has some saturation and compression but I have no idea if that’s just the reality of a raw 909.

Next, I decided to follow Gordon Reid’s “patch” more rigorously by using 3 different voices on the AR.
The first voice used is the DVCO engine to create the kick’s “oomph”.
The second voice is the impulse engine to create the click transient
And the third voice using the noise machine to create the noise transient.
I guess in the traditional 909 architecture, a triangle wave was shaped into a sine wave through a “waveshaper” circuit. I poorly approximated a “waveshaper” using the low-pass filter and I actually liked the resulting patch, but starting with a sine wave sounded more accurate. With some distortion and light compression, this was the patch I was happiest with. Varying the volume of the click and noise transients really changes the character of the overall sound so there is lot to adjust.

Looking at my reference 909 kick’s waveform in a DAW, its clear there was some weird transient shaping going on. My patches have a typical “sloping downward” envelope, but the Divine 909 kick seems to first quickly slope upwards after the initial attack and there is some clipping on the bottom part of the waveform. I tried to approximate this by triggering a LFO on the compressors volume along with adjusting the distortions “Symmetry” parameter, but it seems to give it a different character than the Divine 909 sample.

In conclusion, I don’t feel any of my patches really nailed it when compared to my reference 909 kick, but they are, by definition, 909-style kicks. Here is a link to recorded audio of these kicks (starting with the reference sample).
The kicks in order as they are sequenced in the Soundcloud audio:

1. Divine 909 kick sample (this is the reference for comparison)
2. Hard BD 909 patch
3. Classic BD 909 patch
4. Plastic BD 909 patch
5. Triangle VCO 909 patch
6. Sine VCO 909 patch

I’ve also attached the sysex files for all the patches mentioned here: 909 Kicks for Rytm.zip (784 Bytes)

I would really love some feedback about these particular patches and maybe some clues on how they can be tweaked or processed to more accurately emulate the 909 kick. Please feel free to share your own attempts too!

[loopdude]

quite an effort. but cool stuff you did there.

after listening to the soundcloud examples you posted I would easily say “close enough”. the hard and classic ones are really close. I actually liked them better than the real 909. I will definately download the sysex. thanks for that!

[PeterHanes]

@saint_stink

Please feel free to share your own attempts too!

Some previous attempt(s) here:

Where is my analog TR-909 kit?

I’ve got the “Divine” pack, not impressed with the 64 supplied samples. Is there a synthesized TR-909 kit that sounds convincing? a kit with the ROM extracted untreated TR-909 samples, completed with analog osc sounds would be great.

[loa]

Congrats @saint_stink you did a remarkable work !

[saint_stink]

Thank you very much for the encouraging words, @loa and @loopdude!

@PeterHanes, yes I think this is the exact thread that I was thinking of when undergoing this experiment. Of course, I didn’t think to check it before starting sound design There are quite a lot of clues in here the should help with tweaking my patches.

Since yesterday, I found some more 909 kick samples circulating around the internet and was surprised that:

1. Elektron’s Divine 909 kick that I used as reference is especially bass boosted so mine are a bit boomy as well
2. There is quite a lot of variance in the character between 909’s in general.

I am hoping to take a shot at the snare next.

[loa]

@saint_stink

I am hoping to take a shot at the snare next.

That would be great ! The 909 snare is so pretty !

[loopdude]

Yeah, we need more “clack” in our lives. One thing the rytm lacks a bit imo

[Lucid_NL]

Thanks for sharing, very close! Found this video on Youtube, different approach to the classic sound:

Elektron Analog Rytm MKii Classic BD Demo

[astricii]

Pro tip on snappin up snares on the rytm. which is a bit of info I gleaned from MPC users, set the filter to lowpass bump the resonance up (not to osc.) then slowly back off the cutoff until the peak is in the pocket around that 3k to 6k snap I remember it hitting nice around 110ish? but it’ll depend in your tuning settings etc. This is basically like high resonance high pass trick that people use on Kicks/bass but reversed. the peak filter on the rytm is a great for adding mids emphasis in some use cases too.

[saint_stink]

@Lucid_NL Great tutorial! With this approach, there is a way stronger transient and a much darker tone in general. I really like it and I am sure that it will cut through most mixes. Probably sit real good in a vibey track. Personally, I’d throw in a sampled white noise transient and turn down the sweep depth for less of a “click”, but that’s just the way I like things.

@astricii Thanks, I’ll be sure to try that out! I’ve been using a static (no envelope) peak filter around 106-110ish on a lot of my snares lately so a similar idea, but it might be better practice to use the low-pass when going for classic and vintage sounds.

Eventually, I hope to post what I have for the 909 snare, but so far its rough. Not sure if I have a good enough ear to tune the oscillators just right. Whenever I go back to work on it, I get totally lost and everything starts sounds the fucking same but at the same time completely different?

[Lucid_NL]

Tonight I will use an external envelope (tiptop z4000) via the rytm cv input and try some Jomox Modbase style kicks…

[wald]

Excellent!

Just FYI: your 909 sample is definitely not a cleanly recorded 909. Not sure that matters given the exploration it inspired. But I am happy to send you clean files if you’d like.

[saint_stink]

@Lucid_NL Is the tiptop z4000 quicker than the rytm’s envelopes? I never considered before, but that’s a great solution for doing nonlinear envelopes! I think the CV input is the RYTM’s #1 most underutilized feature.

@wald Yeah, in fact its smashed to hell haha. I wish I realized that before I started And, YES, I would really appreciate it if you could send a clean sample! Thank you!

[grayxr]

I’m obsessed with the 909.

I actually think the BD Hard kick is already almost identical to the 909, but missing that subtle attack noise element. I’d be fine with using a noise sample to bring it closer there. But the snappiness of the envelopes and the raw waveform sound pretty spot on. The 909 circuit uses a diode to clip the triangle VCOs to bring them closer to sine waves.

I’ve tinkered with three different approaches for the snare:

A) Start with the SD Hard machine. Eliminate noise for now, set a pretty decent sweep time / pitch envelope decay, and use a one-shot exponential envelope to get a snappier decay. Then, sample this machine and load it into the current sound’s sample layer. Then tune the sample up around a fifth interval from the machine’s tune and reduce the sample volume a bit so the machine’s partial is more pronounced. Now you can add the noise mix back in.

B) Start with the SD Hard machine. Turn the machine’s decay to 0 so only the noise is present, and turn the noise decay all the way up. Then, sample the noise and load it into the current sound’s sample layer. Now, change the machine to the dual VCO. You can get a pretty decent drumhead sound just by detuning the second VCO about a fifth above (I have detune amt at +30), and by tinkering with osc config (I like the third option), and making the first VCO more pronounced by adjusting the balance. Also set both oscillator decay amounts to 11, and use a one-shot exp LFO on the dual:level param to get a snappier amp env decay. I found it most realistic to simulate a 909 snare “tone” (noise decay) control by adjusting the sample “start” knob instead of “end”, but that seems counter intuitive. Maybe get a few different noise samples and use the LFO instead to cycle through the noise samples so it doesn’t sound so repetitive.

C) Start with the SD Hard machine. Turn the machine’s decay to 0 so only the noise is present. Use a sample of a bare 909 snare (with no snappy/noise) for the “drum”/drumhead sound. Adjust noise level and decay for the snappy mix.

The C) approach to me sounds the most authentic only because an authentic 909 snare drumhead sound is harder to achieve given the current constraints, IMO. The noise on the SD Hard machine actually sounds like the 909’s, though I think the snappy control circuit has a sort of band-pass filter element as well.

[saint_stink]

Thank you so much for this break-down. I am very excited about trying out some of your outlined approaches! I sincerely appreciate this!!

Can you elaborate any more on the diode circuit? Does it simply clip the top of the waveform as straightforward as I imagine it? I figured such hard clipping would add more harmonics than the triangle wave has to begin with. Or does the diode circuit function more like a nonlinear low-pass filter - nonlinearly adds some lower frequency harmonic content but removes the upper overtones that made the wave orginally triangular?

I’ll take a look at an oscilloscope, but maybe driving a triangle too hard into the filter can clip the top of the waveform similar to the diode circuit.

[grayxr]

Does it simply clip the top of the waveform as straightforward as I imagine it? I figured such hard clipping would add more harmonics than the triangle wave has to begin with. Or does the diode circuit function more like a nonlinear low-pass filter - nonlinearly adds some lower frequency harmonic content but removes the upper overtones that made the wave orginally triangular?

Yeah, more the latter – it rounds out the triangle wave. Here’s a deep dive into emulating the 909 circuits using DSP (reaktor). The guy in this video goes over a lot of this stuff with diagrams, etc.

I do wish Elektron would make a dual VCO snare machine with a dedicated detune control. Or if they could expose some tooling in an SDK or otherwise for making custom machines, that’d be even better.

I can try to post some audio samples of my 909 snare tinkering in a bit.

[stankeev]

machine SDK is a dream
at least I want dual VCO with additional parameters (few pages of parameters))), including noise generator, as its separate digital circuit, that unused in DVCO.

[grayxr]

Here are my various 909 snare emulations, denoted by A, B, C per my outline above: https://soundcloud.com/sonic-sand-castles/sets/rytm-909-testing

A = SD Hard machine using a sample of a pitched up bare SD Hard (acting as the second VCO sound / partial)

B (1&2) = Dual VCO machine with sample of SD Hard noise, second version (B2) is slightly more detuned

C = actual bare TR-909 drum sample (with no noise) using the SD Hard machine primarily for the noise mix (0 decay and various noise settings)

Each of these have various AMP and LFO settings, mostly using a one-shot exponential env to modify the level amounts and some shortened decay on the AMP env.

I included some example beats, too.

As you might be able to see, the bare drum sound is hard to achieve given the peculiar timbre of the TR-909’s VCO waveshape. I think Elektron would either need to create a super specific snare machine with that sort of waveshaping, or allow developers to create a custom machine to accomplish it. Or maybe someone else do something clever to achieve a more authentic sound.

[saint_stink]

Once again, thank you for sharing your excellent work and research! Really sounds terrific. I totally hear now how the 909 bare drum-skin sound is very particular. Still happy with your alternatives, since they are both 909-style snares

Re:dual VCO snare machine - doesn’t the “SD Classic” machine have a dual VCO configuration? I don’t have my RYTM in front of me right now, but when I first attempted the 909 snare, that is the machine I selected because it seemed to have the necessary architecture. I recall it even having a detune parameter, but was disappointed that the second oscillator’s decay seemed a bit too short.
My initial assumption was that the both the “Classic” snare and kick engines were an homage to the 909 sound architecture, but I find myself agreeing with your take; the “Hard” engines sound the best in terms of the oscillator’s character.

[saint_stink]

the unused noise generator in the DVCO engine is a damn shame. Including it would really open up a lot of possiblities - maybe as a second page on the SRC screen like how the SAMPLE screen got a second page for displaying the waveform.