I am using 3 AD9833 on a board, with variable reference clock (0-24MHz). They works just fine, when programming sinusoidal waveform.
The communications is with 10x8bit words, in the same FSYNC frame.
Now, the problem arises when trying to program square or triangle, or setting powerdown bits. They simply don't work. The waveform stays the same, sinusoidal, the correct frequency.
I checked the data sent to the chip, by a single trigger on both MCLK and SDATA, MCLK synchronized, and the needed bits are all there, on the falling edge of the clock, where they should be. I am sending the same command at 300msecs, for having MCLK bursts, for easy sync.
Now, the only thing (however, improbable) wich comes in my mind is that the chip can't work if there is some nonuniformity in the clock pulses, because there is a little space betwwen words, in wich time the clock stays high, how it should. In this way, maybe the second byte of the command word is not received correctly.
Now, I know that these chips works, and three faulty chips on a board is not possible.
If somebody have any ideea... many thanks in advance.