ff-ad9833: Added Für Elise Demo

I added this tune to the demo-melodies.fs file:

Für Elise (first half) played with ff-ad9833 code

My apologies to the Beethoven fans as I had to do some violence to the tune in order to make it monophonic and to fit it inside the 3 octaves provided by my code. I also had to apply some filtering to the recording to try to remove background noise in my apartment.

Here is the tune in source code:

create furelise-p1
d_16th ne  o5 == d_16th nd# o5 == d_16th ne  o5 == d_16th nd# o5 ==
d_16th ne  o5 == d_16th nb  o5 == d_16th nd  o5 == d_16th nc  o5 == 
d_8th  na  o5 == d_16th nr  nr == d_16th nc  o4 == d_16th ne  o4 ==
d_16th na  o5 == d_8th  nb  o5 == d_16th nr  nr == d_16th ne  o4 ==
d_16th ng# o4 == d_16th nb  o5 == d_8th  nc  o5 == d_16th nr  nr ==
d_16th ne  o4 == d_16th ne  o5 == d_16th nd# o5 == d_16th ne  o5 ==
d_16th nd# o5 == d_16th ne  o5 == d_16th nb  o5 == d_16th nd  o5 ==
d_16th nc  o5 == d_8th  na  o5 == d_16th nr  nr == d_16th nc  o4 ==
d_16th ne  o4 == d_16th na  o5 == d_8th  nb  o5 == d_16th nr  nr ==
d_16th ne  o4 == d_16th nc  o5 == d_16th nb  o5 == end-score

create furelise-p2 d_quarter na o5 == end-score

create furelise-p3
d_8th  na o5 == d_16th nr nr == d_16th nb o5 == d_16th nc o5 ==
d_16th nd o5 == end-score

create furelise-p4
d_dt-8th ne  o5 == d_16th ng  o4 == d_16th nf  o5 == d_16th ne  o5 ==
d_dt-8th nd  o5 == d_16th nf  o4 == d_16th ne  o5 == d_16th nd  o5 ==
d_dt-8th nc  o5 == d_16th ne  o4 == d_16th nd  o5 == d_16th nc  o5 ==
d_16th   nb  o5 == d_16th ne  o4 == d_16th ne  o5 == d_16th ne  o4 ==
d_16th   ne  o5 == d_16th ne  o4 == d_16th ne  o5 == d_16th ne  o4 ==
d_16th   ne  o5 == d_16th ne  o5 == d_16th ne  o6 == d_16th nf# o4 ==
d_16th   ng  o4 == d_16th nd# o5 == d_16th ne  o5 == d_16th nf# o4 ==
d_16th   ng  o4 == d_16th nd# o5 == d_16th ne  o5 == d_16th nd# o5 ==
d_16th   ne  o5 == d_16th nd# o5 == d_16th ne  o5 == d_16th nb  o5 ==
d_16th   nd  o5 == d_16th nc  o5 == d_16th na  o5 == d_16th ne  o4 ==
d_16th   na  o4 == d_16th nc  o4 == d_16th ne  o4 == d_16th na  o5 ==
d_16th   nb  o5 == d_16th ne  o4 == d_16th ng# o4 == d_16th ne  o4 ==
d_16th   ng# o4 == d_16th nb  o5 == d_16th nc  o5 == d_16th ne  o4 ==
d_16th   na  o4 == d_16th ne  o4 == d_16th ne  o5 == d_16th nd# o5 ==
d_16th   ne  o5 == d_16th nd# o5 == d_16th ne  o5 == d_16th nb  o5 ==
d_16th   nd  o5 == d_16th nc  o5 == d_16th na  o5 == d_16th ne  o4 ==
d_16th   na  o4 == d_16th nc  o4 == d_16th ne  o4 == d_16th na  o5 ==
d_16th   nb  o5 == d_16th ne  o4 == d_16th ng# o4 == d_16th ne  o4 ==
d_16th   nc  o5 == d_16th nb  o5 == end-score

create furelise-p5 d_dt-quarter na o5 == end-score

That occupied 326 bytes of memory, which was the score bytes along with the bytes needed for the word headers (furelise-p1, etc.) It was divided into five parts to save memory, since some of the parts are repeated.

Free software source code and documentation are available here:

https://codeberg.org/infrared/ff-ad9833

Advertisement

ff-ad9833 project: musical notes and melody system implemented

ad9833-based audio module

I added two modules to my ff-ad9833: one called notes which handles playing music and rest notes, and another called score which provides a convenient way to lay out a melody of notes in memory and to play them at an arbitrary tempo. My first simple demo is the “Are You Sleeping” tune:

I defined some Forth words such that it is fairly natural to lay out the notes in a memory block. This is what “Are You Sleeping” looks like in Forth code:

flash

create rusleeping
d_quarter nc o5 == d_quarter nd o5 == d_quarter ne o5 == d_quarter nc o5 ==
d_quarter nc o5 == d_quarter nd o5 == d_quarter ne o5 == d_quarter nc o5 ==
d_quarter ne o5 == d_quarter nf o5 == d_half    ng o5 ==
d_quarter ne o5 == d_quarter nf o5 == d_half    ng o5 ==
d_8th     ng o5 == d_8th     na o6 == d_8th     ng o5 == d_8th     nf o5 ==
d_quarter ne o5 == d_quarter nc o5 ==
d_8th     ng o5 == d_8th     na o6 == d_8th     ng o5 == d_8th     nf o5 ==
d_quarter ne o5 == d_quarter nc o5 ==
d_quarter nc o5 == d_quarter ng o4 == d_half    nc o5 ==
d_quarter nc o5 == d_quarter ng o4 == d_half    nc o5 ==
d_half    nc o5 == d_half    nr nr == d_half    ng o4 == d_half    nr nr ==
d_whole   nc o5 == d_8th     nc o5 == end-score

ram

Each note, including duration, pitch, and octave, is packed automatically into one 16-bit memory cell, so that not a lot of memory is used.

This music system is not really sophisticated enough to use for something like a video game console, since the ad9833 does not have ADSR envelopes or anything like that. But perhaps it would be useful for something like a simple toy, or part of the interface of some appliance.

It has been a lot of fun coding these modules, as well as writing the comment documentation. I developed my own source code documentation standard for the project to keep the code documentation clean and organized.

I learned in the process about the reuse software, which has a nifty reuse lint command which helped me in placing proper licensing and copyright documentation in all my code files. Most free software coders I have come across think very little, if at all, about proper code licensing details. But it is important to make the licensing clear, and clear on all files in the project, so as to make it easier for others to reuse, modify, and share the code with confidence.

While this component of the project is nearly complete, I had thoughts of adding one more demo tune — something longer and more sophisticated. As far as the ff-ad9833 project itself, I had thoughts of adding one more module for generating Audio FSK data communication capabilities. Like, for amateur radio RTTY.

The code for the ff-ad9833 project is available to download from this repository:

https://codeberg.org/infrared/ff-ad9833

In the picture at the top of this post, the audio module is being powered by the Arudino UNO 5V VCC pin. However, it is better instead to power the audio module from a separate 5V power supply (at least 3W). You can then tie the DGND pin on the audio module to a GND pin on the UNO, and also tie the other GND pin on the UNO to the ground or negative lead on the 5V power supply. In my experiments, this resulted in more stable audio module operation, as well as cleaner sound at the higher volume levels.

I think, if I were redesigning my audio module, I might also utilize a diode to prevent current from rushing back from the audio module into the UNO. I’m not quite sure if that is really necessary, but it seems like a good idea.

ff-ad9833 Repo

codeberg repository for ad9833-related FlashForth code

For my on-going exploration of ad9833 audio generation using FlashForth 5, a code repository is now available at codeberg:

https://codeberg.org/infrared/ff-ad9833

So far, I have a demo-pitches word that plays c4, e4, g4, and c5 frequencies, but I can set other frequencies ranging from a4 to g#6, using the words I have so far. There is no system yet for setting duration, so I can’t play notes proper. But it was fun setting up a system for setting the equal temperament pitches.

In microcontroller programming, it is much easier to justify efforts to save memory. I needed to store the precalculated frequency register data values for each pitch. Originally I had a table like so:

$5274 , $4000 ,
$538d , $4000 , 
$54b7 , $4000 ,
$55f2 , $4000 ,
...
$4b5a , $4002 ,

With each line (two 16 bits words) representing one pitch. However, the first two bits in each 16-bit word are actually not frequency data, but register addressing bits, which can easily be added in later. Also, the second 16-bit word in each pair has (in my application) only two bits of actual frequency data, which are the two right most bits. So, I cut the table memory usage in half by packing those two bits into the left-most two bits of the first word in each pair:

\ packaged freq register data for equal temp
\ pitches from A4, A#4, B4 , ... , G#6
\ format numbered from lsb to msb:
\ 0-13: the 14 LSBs of the lsb register load
\ 14-15: the 2 LSBs of the msb register load

create packed-pitches
$1274 , $138d ,
$14b7 , $15f2 ,
$1740 , $18a2 ,
$1a19 , $1ba7 ,
...

Now each line in the table is actually two separate pitches. Then I just need a few words to separate the bits back out into separate words, and recombine them with the frequency register addressing bits:

0  constant na
1  constant na#
2  constant nb
3  constant nc
4  constant nc#
5  constant nd
6  constant nd#
7  constant ne
8  constant nf
9  constant nf#
10 constant ng
11 constant ng#

0 constant o4
1 constant o5
2 constant o6

: pull-pitch ( note octave -- u ) 12 * + cells packed-pitches + @ ;

: pull-14lsb %0011111111111111 and ;

: pull-2msb %1100000000000000 and 14 rshift ;

%0100000000000000 constant FREG0

: tx-pitch ( note octave -- )
    pull-pitch cp>r pull-14lsb FREG0 or 2tx-spi
    r> pull-2msb FREG0 or 2tx-spi ;

In testing, I have some trouble with some pitches getting distorted when I set the volume to higher levels. I suspect the problem is that I am still using the USB power supply and need to feed in my +5V from an external PS. I plan to try that next week, God willing.

SPI Driven Audio Circuit w/ Frequency Selection

Assembled Audio Module based on GY-9833 Frequency Generator.

The PCBs I had ordered from OSH Park arrived, and my co-worker Mike was nice enough to solder the components on for me. The module works, and I was able to produce 1000Hz and 2000Hz audio using FlashForth to drive the SPI signals. The sound output sounds pure and pleasant, although when I twist the volume above halfway, my ears detect what sounds like a very slight overtone or distortion, but it still is pleasant-sounding enough.

The module is a step above the usual cheap piezo buzzer circuit, the latter having the grating, distorted sound which is usually undesirable. But it is a few steps below having an actual programmable sound generator chip or a full blown sound card.

The actual PCB design here would not be used in a commercial product as it is very large for its functionality (about 6 cm wide). But it could be constructed easily by any hobbyist as the components are all through-hole and spaced far apart.

Link to design on OSH Park:

https://oshpark.com/shared_projects/2lO7K5tc

Parts list:

Schematic:

KiCAD files:

These are the trimpots that fit the board:

https://www.amazon.com/gp/product/B071WW6VN8/

The GY-9833 module is also available from several manufacturers through Amazon and other Web stores.

Audio Circuit: Single Frequency Output Selection

Audio Circuit Driven by Op Amp, GY-9833 frequency synthesizer, and Arduino

Above is the circuit I mentioned earlier had been intending to design with a small 3W speaker driven by an LM386 op amp driven by an AD9833 frequency synthesizer chip, driven by SPI (sourced in the schematic from an Arduino Uno).

I thought designing and testing this simple circuit would only take an hour or two. But I got hung up for a lot of hours because, in the breadboard, I was getting weird spurious signals in the output when I had the the circuit set to certain volume levels, and I couldn’t figure out what the problem was. In the end, the problem was fixed by putting a large capacitor across +5V and AGND. After that, this output sounded pure and pleasant, testing with 1000Hz, 1500Hz, and 2000Hz signals. I do not know if it is really necessary to use a capacitor quite that big, but I didn’t feel like doing additional testing.

I had thoughts to turn the circuit into a PCB, though it would be through-hole (easier to assemble) and not a miniature surface-mount design like would be seen in a commercial application. This would be around 6cm x 5cm, with the speaker plugging in on the right pins and the Arduino plugging in on the left header. The GY-9833 module would slide onto the center pins and could be soldered on to them.

I can provide design files later after I finish laying out the traces. I can also post some FlashForth code later.