Hello,
I am building a karaoke / PA sound system with an ADAU1452 and I have a problem that I could not solve after several days of testing. I would appreciate your help.
HARDWARE
- ADAU1452WBCPZ (top marking: ADAU1452WBCPZ #1929 D, Singapore) on a generic development module.
- Crystal 12.288 MHz. PLL: input divider 4, feedback divider 96, PLL locked. I expect a 294.912 MHz core clock.
- Clock generator 1: M = 6, N = 1 (48 kHz). N was 2 until yesterday; changing it made no difference.
- Two CS42448 codecs in 8-channel TDM on serial ports 0 and 1 (ADAU is master). A Bluetooth module (I2S) on serial input 2 through ASRC0.
- Programmer: a DIY USBi clone (Cypress CY7C68013A / FX2LP), I2C.
- Software: SigmaStudio 4.7. Logs before reinstalling show compiler 4.7.0 build 1827; after reinstalling, 4.6.0 build 1809. The problem is the same.
PROBLEM
The project compiles with 0 errors and 0 warnings and downloads normally. But when the total "Inst (cycles)" in compiler_output.log goes above about 3105, the audio of the main chain (loudspeaker, tweeter and subwoofer outputs) stops. This happens even when the added block is tiny (one 6-cycle mixer). Deleting any block, even an unrelated one, brings the audio back.
A short chain (input -> many biquads -> output) always works, even when I multiply the blocks. It looks like the blocks at the end of the program are not being executed.
Above the limit, the 3-way crossover also misbehaves: the Hi output carries mids and some lows, and the Lo and Mid outputs are silent. Below the limit, the same crossover works perfectly. The RMS limiter output is also strongly attenuated.
MEASUREMENTS (from compiler_output.log)
Total cycles / instructions / DM0 / result
2985 / 2973 / 95% / OK
3077 / 3065 / 95% / OK
3100 / 3088 / 95% / OK
3105 / 3093 / 75% / OK
3105 / 3093 / 95% / OK (larger delay)
3106 / 3094 / 95% / NO AUDIO (after adding one 6-cycle mixer)
3112, 3122, 3127, 3273 / more than 3094 / - / NO AUDIO
SigmaStudio reports 51% of 6144 at 3105 cycles. The behaviour looks as if the core had only about half of the instructions per sample (3072 at 48 kHz).
WHAT I HAVE RULED OUT
- Memory: it works with DM0 at 95%. DM1 is at 33-36% and PM at 57%.
- Panic manager: with the failing project loaded, there is no loop overrun, watchdog or parity error. Once, a latched DM0 Bank 1 parity error appeared; it cleared after "Clear panic manager" and did not come back.
- Chip variant (it is not the -150), clock settings, and a clean reinstall of SigmaStudio.
- The Program Counter registers gave inconsistent values, depending on the state of the clear button, so I could not use them.
QUESTIONS
1. Is there any known reason why a program of about 3100 cycles per sample would not finish on an ADAU1452 at 294.912 MHz and 48 kHz? For example, something that uses core cycles that the compiler does not count.
2. How can I measure the real number of cycles executed per sample and the real core clock? How should the Program Counter Length / Max Length registers be read?
3. Can a USBi clone (FX2LP) corrupt the download of larger programs without any error? Do you recommend a test, such as self-boot from EEPROM or an original programmer?
4. Is there a known issue with the RMS limiter block (strong attenuation, very low gain right after download)?
I can send the project file, all compiler logs, and screenshots of the Hardware Configuration tabs.
Thank you very much,
[Your name]