TR909 snare planning

[VA1]

Hi, i want to build a TR909 snare next year.

There is a 2sd1469 in the diy9090 schematic,
what transistor can replace it ?, is it important to the sound ?

And how do you make the noise generator with available parts ?
I dont want to use a MCU for the digital noise, i like to clone the original.

The 4006 is not available, i read somewhere you can buy them in different packages to do exact the same thing ?

And what are those other parts that say "31" ?
Anyone got links ?

thanks

[Electric Druid]

[VA1]

Oh you mean to emulate the exact thing ?, sound like a good plan.
To me its unknown parts, i dont know how it works.

I thought : MCU noisegenerator is just a rand() function to a DAC.

What is the rate from this tr909 noise generator circuit ?, i dont see it.

Oh and by the way : is this noise generator a good project to start with FPGA ?

[Electric Druid]

[VA1]

Ok thank you.

How does the original circuit output the noise on what bitrate ?
How do you output the noise from a PIC ?

[VA1]

[Electric Druid]

[VA1]

Nice, so the noise has 1 bit output.
I did not thought this, i was thinking analog with a DAC.
Then i think it will be good, just shift a 32 bit value and add the XOR of 13 and 31 position ?
Ok.

Very nice it saves alot of parts and space.

[Electric Druid]

[VA1]

Dont i have to initialize the register ?
It handles itself right ?

[VA1]

Oh wait, is the XOR bit taken after the shift or before ?

[VA1]

Ok i finally finished the noise generator in a PIC12F615 running at 8MHz.
I got it running at 3 KHz, i dont know how to make it 100 times faster.
What kind of mad ASM skills do i need ?

Anyways i was playing with this noise generator and you can get very nice noises with changing the feedback-bits.
Freaky roar sounds and beeps + repeating noises.

[VA1]

I calculated wrong, i am running the noise at 12K.
I just increased it to 15K max

If i take a 32MHz chip i can get 60K max.

[Electric Druid]

With the chip running at 8MHz, you've only got a 2MHz instruction cycle, so that doesn't give you much time. Say we're aiming for 100KHz output rate - that's only twenty instructions. Still, for a basic LFSR in assembly, that's possible.

How long is your code?

Here's an example (this was done on a 12F675)

[VA1]

I am making a new project now i spend the whole night yesterday : UART controlled noise generator.
The xor bits can be selected and the pitch can be set.

I am making all options available, taking xor from all combinations possible,
i dont need the high beep tones, the roaring electric beep sounds i save with the noises.
There are many noises that has tones hidden in it, small soft glitches, have to sort them out later.

What a fun you can have with a small chip.
I go get the inner core code from my workcomputer for you now, wait a minute.

[VA1]

union ShiftReg{
unsigned long value;

struct{
unsigned b10:1, b11:1, b12:1, b13:1, b14:1, b15:1, b16:1, b17:1;
unsigned b20:1, b21:1, b22:1, b23:1, b24:1, b25:1, b26:1, b27:1;
unsigned b30:1, b31:1, b32:1, b33:1, b34:1, b35:1, b36:1, b37:1;
unsigned b40:1, b41:1, b42:1, b43:1, b44:1, b45:1, b46:1, b47:1;
};
}shiftreg;

void init()
{
shiftreg.value = -1;
}

for( ;; )
{
if( TMR1IF )
{
TMR1ON = 0;
TMR1IF = 0;
TMR1 = -( 33 + pitch.frequency );
GIE = 0;
TMR1ON = 1;

PORTA = shiftreg.byte1;

feedback = shiftreg.b47;
feedback ^= shiftreg.b25;

shiftreg.value <<= 1;
shiftreg.b10 = feedback;
GIE = 1;
}
}
}

[VA1]

Oh the GIE has to be before setting the timer btw..

[Electric Druid]

It looks pretty efficient to me (at least for C!).

The only way you'd manage to get that to go faster would be to have a look at what it compiles into, and then see if altering things makes the resulting assembly longer or shorter. Is it worth it? Only you can decide.

I don't know enough about how the compiler assembles things like your Struct, but it's possible there would be a simpler way to get specific bits there.
That said, your way is *variable*, which is generally more costly in terms of time than a single specific situation, where you can fine-tune the code for just that situation.

Often I've chosen tap sequences based on where in the byte the feedback bits fall in order to minimise the bitshifting required to XOR them together. For example, if both bits were the 6th bit in the byte, you could XOR the two bytes directly and then use the 6th bit as the result. If one bit was the 4th bit in the byte and the other was the 6th, you'd have to shift one of them twice (or shift both of them once) to get them to line up before you could XOR them. Obviously, there are lots of ways you can do this, but this gives you the general idea - a specific situation can be heavily optimised, a general situation not so much.

[VA1]

I know i want to build the hardware version one day for the drums.

This noise PIC project i could spend a whole year, i still dont have perfect noise without glitches.

What would the HW version reset its bits to ?, random at startup ?

I made a LUT of 128 choises for the reset value, it is not enough.
Now i have to build a edit function for all 32 bits and i gave up.

It has some nice drums is pitch envelop is used, retrigger also gives nice sounds.
I like to make a single OTA stereo panning thing for these noises.
Together with a super small VCF, the smallest i can build.
Need to add another small chip for LFO and ENV.

I better spend time perfecting my already build synth.

[Electric Druid]

[JovianPyx]

This subject has been at least on SDIY, that is, the idea of getting glitch free noise from an LFSR (this is a shift register type pseudorandom noise generator). The more bits and the faster you run it, the higher the quality of noise will be. Personally, I use no fewer than 32 bits and I get very clean noise at normal audio sample rates. However, I also use fast microprocessors (STM32F and STM32H) which will do it without breaking a sweat.

There are also techniques that can be used with less bits to make the stream more random. This amounts to using two LFSR registers of different lengths and then you can do things like interleave the bits. For example, an 8 bit and a 7 bit LFSR could potentially be as good or better than a 15 bit register. The drawback is that like using a longer register, the fact that you're doing two LFSR actions for each new random number, it is slower than just using a single 8 bit register which as you know can sound like chuff-chuff-chuff...

[Moosgummi]

Oh wow,
I had a plan to build the coolest CMOS LFSR ever with 159 bits, using a 128 bit SR (probably 4562, my favorite CD4*** site disappeared, and my CMOS box is just a box with bags and numbers) and a programmable SR (4557?) set to 31 bits.
But now I see just the 31 bits alone give me 2 hours of noise @ 300kHz, so that seems excessive...
Cool nonetheless!

[CHRISKELLY]

Not sure if this will help, but this article gives a good CMOS chip configuration and also explains how to maximise the sequence length with various tap positions

https://www.digikey.co.uk/en/articles/techzone/2018/mar/use-readily-available-components-generate-binary-sequences-white-noise