Cetone Development Blog

2008-07-14T01:14:00.004+02:00

About my YM3812 emu ...

well, I've read a lot of stuff about the yamaha chips, and also some patents.
Some patents also contained some schematics ... and while thinking about them, I solved
a small problem with some sounds in my emu.
There has been many stuff about the Hunting Phenomena and how they fixed it (using a 32 step shift register in an averaging circuit) ... this should work like a low-pass. (They seem to use this shift-register-averaging circuit on everything: operator outputs, feedback, phase modulation input)
So I've implemented another (right now integer) leaky-integrator inside the PM path and voila: sounds just like my AWE64^^

2008-07-13T12:51:00.010+02:00

Ok, system runs stable, Win and SuSE are configured again. No data loss.

I've done an emulation of the YM3812 during the last days (melody mode only). This will be used in my Dune2 remake, to play the original .ADL sound files directly from the .PAK archives. (The .ADL is also ready, and I play those old Dune2 ingame tunes over and over again :) (linux, portaudio (and jack))

Some infos about my emulation:
I'm not doing bit-correct YM3812 emulation, but sound-alike emulation. I'm using a 21 bit phase counter, 16 bits fine sine-tables, but YM3812 volume calculation (well, I don't use this strange floating-point implementation, but 12 bits fixed point).
My EG implementation also differs from the one seen in fmopl.c. I don't use those strange step-tables, but a 24 bit EG counter and simple integer-step values -> (4+(0 to 3)) << (rate - 1).
Another major difference to the fmopl.c YM emu is: My emu always runs at the real chip timings. Sound output is done by 'sampling' the emu's output. (In fact: a ring-buffer and linear interpolation). So my emulation always sounds identical, no matter what samplerate you use (fmopl.c sounds different (and strange) depending on the samplerate). Well, a(n) (dis-)advantage of this method is: The cpu consumption doesn't change with samplerate.^^
Finally a 12.7kHz lossy-integrator 24dB low-pass is applied.
I've checked the output of my emu against the recordings of the Dune2 music, done using my old Soundblaster AWE64 Gold: Output is close enough to the original.
(Here's the source, if you want to take a look (still under development) neo_ym3812.cpp neo_ym3812.h )
And now I've got an idea: I'll write a YM3812 VSTi ... a simple one channel version, polyphonic. (My full nine channel emulation uses about 2% to 3% cpu on my 3.2GHz P4, so the VSTi will also have good performance.)
It will contain no anti-aliasing stuff, just pure aliasing-like-hell, retro PM sound.

But first, the final version of CetoneSynth2 will be relased ... and maybe a GUI for CetoneOrg ... but I've still no idea how it should look.

Stay tuned^^

2008-06-17T11:17:00.002+02:00

Another BSOD ... those Corsair RAMs are strange^^

Well, SPD info was incorrect, BIOS reported a CAS delay of 2.5, my RAMs have 2.0 ... changed that.

*sigh*

I hope that my system runs stable now^^

2008-06-15T17:18:00.004+02:00

Well, system is up and working.

-> memtest revealed a bad RAM stick ...

Below a picture of the cause-of-all-evil:

Below: Neopc beta

I still have a lot of work todo ... but everything's working.

Two strange things: WinXP installation hehaved different (again) and I got two BAD_POOL_CALLER bluescreens ... well, I raised the DIMM voltage ... seemed to help.

2008-06-13T17:30:00.002+02:00

Another update:

I tried to recover my HDDs using Knoppix 3.4 & 3.7 no success, Ubuntu 7.04 also.

Then I tried to install a rescue-winxp on an IDE HDD ... no success (maybe all my old IDEs are really out of order, dunno)

Now, what worked:
- reseted all BIOS settings to their fail-safe defaults
- disabled hyper threading
- removed all RAMs, inserted a new one
- disabled my on-chip SATA , enabled on-board SATA
- low-level-format of a SATA HDD which seemed to have errors last time -> now worked
- installed a rescue-xp on this SATA drive

Well, currently I'm copying about 150GiB date to my wife's pc ... looks good.

After that I will low-level format all my SATAs and reinstall everything ... WinXP and openSUSE

... this will take a while^^

2008-06-13T00:37:00.002+02:00

*wuuuuuargh*

linux also crashes

well, two things could cause this: general hardware failure, or just crashed sectors on the hdd ... dunno ...

... will go to sleep now

tomorrow i'll check the bios, reset all config values, boot from a live-cd and check again

_ D A M N _

*sigh*

2008-06-12T23:43:00.001+02:00

Operation NEOPC Rescue update:

C# and C++ folders are currently transfered to my wife's pc ... looks good

2008-06-12T23:31:00.002+02:00

*sigh*
... damn

I've killed windoze today ... and also messed up my HDD partitions ... 320GiB scrambled NTFS ... at least windoze doesn't like those ...

... luckily my main OS is openSuse right now^^

yet, damn

I will try to recover as much as possible, using my wife's windoze machine (_now_ the 250GiB IDE HDD which I gave her can do its best)

Nevertheless it will take a while until I am able to boot windoze again ... and some more days to re-install all the software.

*sigh*

Hmm ... just the right time, to redesign the partitions and make a clean win/nix install.

*sigh*

2008-06-12T09:22:00.001+02:00

PS: I've noticed that I've forgotten to change the CS2 VSTi version number inside the .dll and on the GUI.
Still it _is_ v0.7.

2008-06-12T00:50:00.002+02:00

So, a few emails in my inbox (and my wife) made me do a minor (or major) update of CetoneOrg and CetoneSynth2.

Changes in CetoneOrg (12th June 2008):

Fixed MSVCRT80.DLL issues (should now run on every Windoze machine)

Changes in CetoneSynth2 v0.7:

Fixed MSVCRT80.DLL issues (should now run on every Windoze machine)
Added VST parameters for MIDI controllers (Ableton Live 7 now shows 46 automation parameters)

Still more to come ...

Good night and good luck^^

2008-04-08T09:11:00.002+02:00

Well, well ... it's been a long time. Many things have happened during the last months and I didn't really had time to work on my plugins.

But the development is not stopped and I think that I can continue work in a few weeks.

So ... stay tuned.

Yours

Neotec

2008-02-11T00:56:00.000+01:00

So, I took a break from working on CetoneSynth2 and made a small ... let's call it 'fun' plugin.

I allways wanted to know how a Hammond sounds like ... or better: I allways wanted to have a small, free Hammond like VSTi. After querying google about "freeware hammond vst" didn't give any usefull results, I decided to do my own "Hammond like VSTi".

Because I don't own a Hammond, I looked up into the Hammond patent files, Wikipedia and other resources. After some thinking I came up with this: "CetoneOrg"

1. I implemented 89 tonewheels, using the original frequency dividers given in the patent file and 60Hz AC (virtually).

2. step was to figure out, how these tonewheels are controlled. After some more thinking I used following aproach:
All tonewheels are running constantly and when a key is pressed there are only "mixer control voltages" sent. "Higher voltages" override "lower voltages".

3. Harmonic foldback ... that was easy.

4. Key click ... was easy too.

5. I heard that when you press a key on the Hammond slowly, the tonewheels get "switched on" in series with a small delay ... so I implemented this controlled by velocity. Dunno if a real Hammond sounds like this, but I like it.

6. I thought about a virtual Leslie but discared this idea because I didn't want to simulate a doppler effect. So I just implemented a LFO which controls vibrato and panning and made the phase of the panning modulation adjustable.

7. Added a chorus/flanger

8. Added reverb

9. Added some controls for key click amount and tone (changes are subtle and are more audible when using higher FS)

10. Added pre and post volume for saturation

11. Added saturation to the chorus unit

12. Added a final low-pass filter

Now, some specs:

- 89 voice polyphonic (61 notes)
- Adjustable tone and key click
- LFO for vibrato and panning modulation
- Chorus/flanger
- Reverb
- tanh like saturation
- no GUI (right now)
- no presets (right now)

I will upload this thing tonight ... maybe someone will use this ... and if so, start a thread at KVR and notify me about this.

2008-02-08T19:15:00.001+01:00

Ok, bug fixed v0.6 is out.

This is hopefully the last beta release^^

2008-02-07T22:42:00.001+01:00

Just to explain the fix I'll apply: There's some unwanted noise when using super mode and morph modulation. I'll remove this ...

2008-02-07T20:20:00.000+01:00

Well, this update will take a little longer, because I have to do some major OSC fixes. I'll have to try something and I hope that I can upload the new version this evening.

2008-02-07T18:33:00.000+01:00

I have discovered a BUG. Found out that I mixed up two variables which caused the main panning be overwritten by the audio input panning.
I'm going to have a smoke and a coffee ... after that I'll upload another fix ... v0.56 ...

I wonder how many 'fixes' I'm going to upload until CS2 runs flawless.

Yours

neotec

2008-02-07T08:49:00.000+01:00

TODOs:
- Change voice allocator to search for voice with lowest volume
- Implement standard VST parameters for easy automation
- (maybe) exchange OSC core
- still add more presets
- Reduce GUI CPU load on changing parameters using MIDI CCs

I've uploaded a bugfixed version (v0.55) yesterday to fix some problems with Cubase and Savihost.

2008-02-02T20:33:00.000+01:00

v0.5 is released

2008-01-31T22:22:00.000+01:00

The ability to change the gui made it neccessary to implement a simple splash screen.

This one only shows the synth's name, current version and copyright for 2 seconds. And only on first load and open.

Sorry, but I have to do this :)

2008-01-31T21:04:00.000+01:00

I finally decided how to use the free space in the lower right corner of the gui: I added an "Audio Input" panel.

Now you can use the sidechain as a modulation source and you may route it through the synth.

It is also possible, to use 2 different signals, e.g. a modulation signal on the left channel and an audio source on the right channel.

AudioVolume and AudioPanning are added to modulation destinations.

I think this is a good solution ... and also sounds very interesting^^

2008-01-29T21:48:00.001+01:00

Next version will be v0.5 (comming soon).

Changes since v0.4:

BUGFIX: Filter overdrive caused explode protection to reset the filter
DESIGN: Gui changed, gui.dat added
IMPROVEMENT: Reduced alias to nearly zero on morph sweeps
IMPROVEMENT: Reduced alias on AM modulation (slightly).
PRESETS: Added some more evaluation presets. (Still not the final ones)
MIDI: More midi CCs implemented

I think I'll test this new version for a few days and release it on friday.

2008-01-29T09:16:00.000+01:00

I am now focusing on reducing the aliasing of CS2 when using sync or AM. May the triangle waveform will lose it's morphing abilities.

2008-01-28T23:51:00.000+01:00

Today I excluded the gui resources from the plugin dll. CetoneSynth2 can now be run with different skins (which of course must be created first). Simply exchange the 'gui.dat' file. C# tool for creation is done: Text file describing sizes and positions, some bmp files. Start the tool, copy the .dat, ready.

Have to go to bed early today ... cu.

2008-01-28T12:13:00.000+01:00

Well, well, there has been nothing wrong with my old formulas to calculate the sync freq ... this one is nicer though. The aliasing when using sync has some other source: The hard edges when restarting the oscillator.

I'm checking on how to smooth those edges without having to double (or quad or ...) sample the oscillators.

Sync must be considered 'experimental' at the moment.

CHANGED: Mod6 amount mod dest behaviour.

2008-01-28T09:43:00.001+01:00

Today I'll revise the oscillator code in CetoneSynth2. Sync still does produce lots of aliasing when using with high frequencies. After thinking about this (when I was trying to find get some sleep) I've come up with a simple equation to calculate the frequency of the slave oscillator (modulated by the master) for anti aliasing:

slave-frequency + master-frequency.

I'll test this today and also give the oscillator a wider frequency range (note range stays (0-120)) to avoid aliasing.

Man am I tired ... *sigh* ... fell asleep at about 0430h ... had to get up at 0800h but slept 'til 0915h ... so I'm already late for my 'job' (I'm currently unemployed and have to do some job with <1$ payment per hour (forced by our "agency for work" which gives us money because we don't have a job)).

2008-01-27T23:07:00.001+01:00

I've found another bug: When using full gains on oscillators and filter the filter-explode-security-device resets the filter. This results in high frequency pulse output.

Found and fixed this one. Will be included in the next release (v0.4 still has this bug).

2008-01-26T23:12:00.001+01:00

The synth engine won't change anymore, I hope.

So, CC MIDI input and output is next.

2008-01-26T23:01:00.000+01:00

Updated list of modulation sources and destinations:

Sources:

Velocity (uni)
Keytrack (bi)
LFO1 (bi)
LFO2 (bi)
Envelope 1 (uni)
Envelope 2 (uni)
LFO1 x Envelope 1 (bi)
LFO1 x LFO2 (bi)
Modwheel (uni)
Modwheel x LFO2 (bi)
Aftertouch (uni)
Left audio input (bi)
Right audio input (bi)

Destinations:

Filter cutoff
Filter resonance
Filter gain
Main volume
Panning
OSC 1 volume
OSC 2 volume
OSC 3 volume
OSC 1 pitch
OSC 2 pitch
OSC 3 pitch
OSC 1 morph
OSC 2 morph
OSC 3 morph
OSC 1 spread
OSC 2 spread
OSC 3 spread
LFO 1 speed
LFO 2 speed
Main pitch
Modslot 6 amount

2008-01-26T02:12:00.000+01:00

I decided to implement fix no. 2. Doesn't make any great difference in speed. (Thank God I've done all this interpolation stuff reusable and already optimized the coefficents calculation inside the filter.)

Now the filter doesn't explode (with low Q settings it never explodes) but the filter still produces some heavy gains when using high Q, very low FC and deep notes. But this may be considered a feature not a bug^^

I'll now think about this gain thing.

have a good night ...

2008-01-26T01:47:00.001+01:00

I've noticed something and that's really bad ... *sigh*

My favourite filter design becomes unstable when quickly changing from greater to lower cutoff frequencies and explodes. The beta version already had a safety condition which resets the filter when it's going to explode (you'll hear a sharp click noise).
After tinking about this for a while I came up with some different kinds of fixes:

1. Limit the minimum FC. But this limit prevents the filter from running nearly closed, so this is not the perfect solution.

2. Always interpolate FC inside the filter. This should prevent the filter from exploding but would also cause some more CPU load. I can't say how much this will be, but I'll check this.

3. Smart interpolation of FC. This means: checking for rapid FC changes towards zero and only interpolating in this cases ... but this would cause massive pops in CPU usage when 'going down'.

4. Exchange the filter. <-- never ever.

Well, it's late at night, I have to go to bed but I will think about this and try some solutions tomorrow.

Stay tuned ...

2008-01-26T00:58:00.000+01:00

BUGFIX: Arpeggiator now starts at note 1. (Before this fix it played the 'real' note first)

ADDED: MIDI program changes

2008-01-25T23:49:00.001+01:00

ADDED: ModSource Modwheel*LFO2

2008-01-25T23:22:00.000+01:00

BUGFIX: Note-off wasn't handled properly in mono mode.

ADDED: ModSources: Modwheel and Aftertouch

CC implementation started (this is sooo easy because of the synth's architecture) ... still thinking about which parameters will be controllable and wich CC should get assigned to it.

TOY: Knobs react on CC input when GUI is open.

2008-01-25T19:17:00.001+01:00

The beta is released! Get it here.

2008-01-25T17:49:00.000+01:00

Checklist:

[x] Chaos
[x] Interpolation stuff for all volume/panning related stuff
[ ] All mod destinations (2 are missing right now)
[x] Architecture select tool
[x] Clean Non-SSE/2, SSE and SSE2 builds
[x] Some test presets

So ... maybe ... maybe ... there's a wolrd where we don't have to ruun ... nope, maybe I might release the beta tonight.

2008-01-24T21:34:00.001+01:00

Modslots are implemented, glide is working.

Yet modslots are still missing some routings ... but I think I really can release a bet version this saturday.

2008-01-24T15:35:00.000+01:00

CS2 now shouldn't crash when using a FS below 44.1kHz.

Things missing (for the beta release) right now are:
1. Glide
2. ModSlots
3. Chaos

I'm getting close^^

2008-01-24T01:18:00.000+01:00

Oh ... I forgot something: CetoneSynth2 currently doesn't support standard VST parameters, so the non-GUI editor won't work.

I dunno if I ever include standard parameter stuff, but I will make most of the parameters MIDI controllable.

And now: Again "have good night"^^

2008-01-24T00:53:00.000+01:00

Before I _have_ to take a nap, here a picture of the CetoneSynth2 beta GUI.

There's still some space left in the lower right corner, which definetely will get filled until the final release.

And, yes, LFO frames are not yet finished. Also modslots are still missing.

Another 'and': This plugin is designed for a FS of at least 44.1kHz ... anything below may currently crash the plugin and/or produce garbage. (But this might not be a real problem)

Good night and good luck ...

2008-01-23T20:11:00.000+01:00

Discovered a major bug in MIDI handling ... *sigh* ... corrected^^
"autoload.fxb" added, for easy preset updates or user defined banks.

"autoload.fxb" must be a complete bank, i.e. 128 patches.

Switched filters from 32bit floats to 64bit doubles. CS2 will come in 3 flavours I think: Normal, SSE and SSE2 compiled.

CS2 should work with: Logic (look here), Cantabile Light, FL Studio 7 demo, Cubasis InWired, Reaper freeware ... and I hope many, many more^^

So ... "Heroes" is on TV ... after that: Stargate Atlantis ... cu^^

2008-01-22T20:46:00.000+01:00

Another sound sample:
LPF sweep, Arp using 128th notes syned to 120BPM

c64.zip

2008-01-22T14:21:00.000+01:00

New GUI shot (using cantabile) -> working3.png

2008-01-22T12:01:00.001+01:00

I've totally rewritten the GUI to Synth parameter handling after I realized that the Synth engine doesn't has to have functions for making the GUI reading or writing values more easily.
Now the Synth engine only fas functions for setting parameters and only one or two (needed) references to the GUI.

GUI GDI object count should be approx. 60 or 70.
Memory usage is approx. 6MB (when using more instances of CetoneSynth2 the amount of used memory increases slightly (a few KB or so per instance)).

ARP sync is done at sample accuracy (I'll implement sub-sample accuracy today).

Again: reduced CPU usage.
I've tested the plugin on my wife's AMD Duron 1GHz, 16 voices, 3xSuper, 24dB makes a total of 50% CPU usage.
Tests on an AMD Duo 2.2GHz, same sound, about 25% usage.

I think the CPU usage is acceptable.

Stay tuned: Beta version will hopefully be available on the 26th January.

Some words about that beta version: I need your help^^
So, please, all of you, liking my synths, please test CetoneSynth2 in your VST hosts and report bugs to me. This time I want to release a (more or less) perfect product including a real documentation.
And _you_ can help me.

The beta release will be anounced at KVR Audio's Instrument forum.

2008-01-21T16:54:00.001+01:00

Currently I am testing the arpeggiator ... maaan ... this is retro chip style^^

*sigh* ... I could just play along forever^^

2008-01-20T13:25:00.000+01:00

Some facts about the voice management of CetoneSynth2:

16 voice polyphonic, each voice having:
3 OSCs (depending on mode up to 15 OSCs)
1 filter
3 envelopes
2 LFOs
1 ARP

Voice allocation rule:
1. search a voice that is done, if found, use this
2. search the 'oldest' voice in release state, if found, use this
3. use the 'oldest' voice

2008-01-20T02:49:00.000+01:00

After some slight optimizations does my synth run at an average cpu load. When using 4 voices the cpu usage is about 10%, when using 16 voices 30% and more. (Test sound: 2xNorm-OSC, 1xSuper-OSC, 24dB filters with 'Exp' envelope mode)

Some words about my 'machine':
3.2GHz P4 HT, 1.5GB Corsair DDR2/400 RAM, WinXP, no ASIO

Well, I'll do my best to make this thing as low on cpu usage as possible (limited to my possibilities of course^^), but this is not Synth1 (I tried this thing yesterday, damn nice, damn good coded).

The mass of options make this synth really hard to optimize. I could get some speed gains by rewriting highly frequented portions in assembly, but my assembly days are long over (back in 1994 I used assembly a lot).

Branches do kill the cache and I try to avoid calls und branches where ever this is possible. I thought about writing different voice render functions for different osc and filter configs, but this would be overkill (pow(4, 3) * 2 functions would be needed).

So, CetoneSynth2 will be a synth which needs a somehow up to date machine if you intend to use a huge amount of voices.

Ah ... GUI has changed again^^

The MIDI sync stuff is developed (I hope) and now only needs to be coded. LFOs have to be implemented (easy) and the modulations must be designed (types of sources and destinations, destination depending amounts).

The last parts will be the arpeggiator, glide, bank and preset handling.

There may be a beta version out next weekend, depdending on how much time I'll have next week.

2008-01-19T14:00:00.000+01:00

Another screenshot showing work in progress: working2.png

2008-01-18T17:15:00.000+01:00

Here's a simple test sound, to let you hear what this synth will sound like: saw_pad3.zip (1.2MB).

Some infos about this sound:

3 oscillators operating in "super" mode with a spread value of 2
OSC tuning: 0, -12, +12 semi
Oscillator volume causes mixer to clip.
12dB lowpass filter with envelope mod.

I like it^^

2008-01-18T16:35:00.001+01:00

The "super" mode now uses only 5 oscillators. Fat enough, less cpu usage.

2008-01-18T15:32:00.000+01:00

I'll do my best to optimize this synth, but the "super" mode is a cpu killer^^

Tests using my synth in cantabile, 3 activated "super" oscillators and 32 voices result in a cpu usage of about 80% on my 3.2GHz P4 HT.

But, hey, who _really_ needs 32 voices of the same sound?!

When using realistic 4 or 8 the cpu usage is ok, but could be better (as said above, optimizations will follow).

Personal opinion: I love the sound of this synth.

Summary: "super" is (and will be) a cpu killer, use it wisely^^

2008-01-18T00:18:00.000+01:00

The "super" oscillator mode is ready, first tests are running (only OSC and ENV), sounds nice.

Explanation: "super" is a homage to the classic Supersaw. When a CS2 OSC is in super mode, it is in fact 7 oscillators, slightly detuned. One interesting fact is, that this "super" mode is not limited to saw waves only.

The wave-detune spreading is adjustable (due to a lack of size in the OSC frame, this is only a small slider).

This makes the OSC count of CS2 a total of 21.

2008-01-17T17:46:00.000+01:00

Ok, I think the oscillators are ready now.

Oscillator range is MIDI notes [0, 120[ ([C-2, C8[)
Real frequency range is [8.176 Hz, 8372.0 Hz]

New mode added: "Super" (guess what this means^^)

2008-01-17T14:58:00.000+01:00

Today I'll implement the buttons. Most of the code needed for those is done.

Again the GUI layout has to be changed (filter frame).

Oh, some remarks to my GUIs.

I am no graphic artist, as you may have guessed already, I am a coder, and so do my GUIs look like. The images are all 'programmed' not painted. I write a (more or less) small program in C# and code all elements. Makes some stuff easier^^

2008-01-17T09:15:00.001+01:00

Here's a first shot of the GUI with all knobs implemented: working.png

The mouse wheel crusade

2008-01-16T19:41:00.000+01:00

Finally it's done. After two days of google'ing, researching and trying I have found the solution:

For all you VST GUI coders (I don't mean VSTGui, but your own windows GUI) that want to use the mouse wheel, just add SetFocus(hwnd) into your WM_xBUTTONDOWN handling.

So, now you can change the value of a knob by simply moving the mouse over it an turning the mousewheel. If the knob won't turn, just click once into the GUI (anywhere).

*sigh*

2008-01-16T14:43:00.000+01:00

So, new GUI preview image online. OSC frames changed. See here.

Update

2008-01-16T10:48:00.000+01:00

Finally I am working again on my plugins. Many things have changed during the last months and I was really busy doing some other coding stuff, but now I really want to revise my plugins (right after I have finished CetoneSynth2).

One really important thing I forgot when designing my first plugins was to compensate for denormalized numbers. This will be fixed.

But now some text about CetoneSynth2:

I started to work on this new synthesizer about four days ago. After two days I already had the GUI partially working (knobs).
After some more thinking about this synth I can now present a list of features quirks:

The Oscillators:
The OSCs are using standard waveforms (bandlimited of course) saw, triangle, pulse and noise. The noise waveform will have some kind of pitch ability.

Semitone range from -64 to 63
Finetune range from -64 to 63 cents
Morph controls either the pulse waveform's pulse width or the triangle waveform's sawtooth-to-triangle morph parameter.

There will be a nice feature, but I'm not sure if it is really usable: Different oscillator modes.

'Normal' mode:
The oscillator works like a standard analogue one, producing one waveform.

'Add' mode:
You don't select only one waveform at a time but you may select a combination of waveforms which then is mixed and outputted.

'Mul' mode:
Same as 'Add' mode, but with waveform multiplication (AM).

I'm really curious about how those OSCs will sound (early sound tests produced nice results).

The OSC frequency range will have a maximum at about 9kHz I think.

Later more ...