Hello keddy,
A measured ADAU1701 EMI spectrum appears below:
We see the crystal’s 12.288 MHz fundamental frequency at A, but virtually no second or third harmonic at B and C respectively. That’s because the crystal’s waveforms are nearly sinusoidal, not rich in harmonics. There’s a huge blip at 49.15 MHz (D) – this is the MIPS rate at which this DSP is performing its calculations (1024 instructions x 48K samples per second). Lots of internal transistors switch at this rate, generating noise at this frequency and its second harmonic (E) at 98.3 MHz. This signal is strong enough that sometimes I use a FM radio to find out if the DSP is running! What can be done about it?
- Skfir’s suggestion to apply shielding makes sense. Likely the RF circuits in your FM receiver are already shielded; placing a metal shield (such as those commonly used with ARM processor ICs) over the -1401 and surrounding components would help.
- SIgmaDSPs produce less EMI to begin with when the PCB layout carefully follows the Best Practices advice at http://ez.analog.com/docs/DOC-1889. A comparison of two prototypes at http://ez.analog.com/message/57200#57200 illustrates the wisdom of observing these recommendations.
- It may be possible to move the second MIPS harmonic out of the FM band as described next:
Assuming your PLL is set for the 256xFs mode, substitute for the usual 12.288 MHz crystal, a 13.56 MHz crystal -- this slightly overclocks the DSP to a MIPS rate of 54.24 MHz. Its second harmonic, 108.48 MHz, is now out of your way. According to the data sheet, a MCLK rate of 13.56 MHz, corresponding to a period of 73.7nS, just squeaks by the rated 73nS minimum period:
This crystal change will have your -1401 actually running at a sample rate of 52.97 KHz instead of 48 KHz. To compensate for this, reduce the set frequencies of filters, etc. in your project, multiplying them by 0.9062. Finally, adjust your ADC input resistors following Page 20 of the ADAU1401 data sheet.
You could instead decrease the crystal frequency enough to slide that second harmonic below 87.5 MHz – this underclocks the DSP, so there’s much more margin in this direction. Going this way, you may need to choose the 2x program length (about 500 instructions max at nominally 96KHz) in order to keep your actual sample rate high enough to cover the full 20KHz audio range.
Best regards,
Bob