Multiplication doesn't work
Thread Summary
The user encountered issues displaying 28.0 format values in a readback block, which only supports 24.0 format and truncates the lower 4 bits. The final answer suggests using a pair of fixed-gain blocks with a gain of 4 to shift the value left by 4 bits, ensuring it is not lost during truncation. The accompanying answer clarifies that multiplying by 1.0 in the standard 8.24 audio format returns the same number, and recommends changing the DC source format to 5.23 for better results.AI Generated Content
Hi,
I just read the article about number formats in Sigma DSP and also about the readback function, which can only display a 24.0 number format and thus truncates the last 4 bits of the input value. So I thought to display a value in 28.0 format I simply have to multiplay it by 16 to shift it by 4 bits to the left so I receive the correct value in the readback block. But even the multiplication itself doesn't work.
In this example I'm just multiplying the number 16 by 1, so basically don't alter it at all. But no matter what values I put in any of the two blocks anyway, the readback always shows 0.000000000. Why is that?
When I connect the constant value directly to the readback block, it gives me the answer as expected:
Edit Notes
slight edit for better understanding
[edited by: the-powl at 9:53 AM (GMT -5) on 12 Nov 2022]
software
number format
readback
Hello,
By definition, multiplying by 1.0 in the DSP's standard audio format (8.24 for the -1452 and newer, 5.23 for the older models) returns the same number. An integer 1 represents the lowest bit in the register, so multiplying by integer 1 gives a vanishingly small result. Try changing DC Source 1_5 in your screenshot above to 5.23 format.
Best regards,
Bob
I think this is due to the difference in size between the readback registers (24-bits) and the standard audio format size (28 bits). When copying to the readback registers, it truncates the lower 4 bits. So any DC value in 28.0 format is read as 1/16th the value by the readback.
In your "gives me the answer as expected" example, you can see the value of 16 (0x10) is 1.0000, or 0x00, 00, 01, as the bits are shifted right by 4. Elsewhere I've seen a recommendation to use a pair of fixed-gain blocks (with a value of 4) to increase the level to a point where it won't be lost when moved to the readback register. I'm using this technique (screenshot below) to get the value of the index going to an index-based crossover for a powered speaker project that I'm working on.
Please log in to reply.