Decimation and filter BW's of the 5660

[cochenob]

A few quick questions regarding decimation and filter BW's in the 5660....

1. I'm sure there's an obvious reason for this....but why are the decimation factors limited to powers of 2?  i.e.  N=32, 64, 128, etc..

2. Attached below are some 'scans' of the 5660 with different BW settings.  For example, in the picture "BW-200kHz...250KSamp_s"....the BW is set to 200kHz...which results in a 250KS/s rate.  I then slowly swept my input frequency over the expected BW (using a 10MHz carrier in this case).  With the output IQ samples...one can then calculate magnitude (white plot, left axis) as well as frequency (after doing a derivative on the phase...red plot, right axis).  This allows me to "map out" the actual magnitude response of the DDC filters.

Basically...I'm seeing that the measured BW seems to be about 10-20% larger than advertised.  So for the 200KHz setting...I measure ~240kHz, etc..

Just curious if this is to be expected...and if the BW's that are advertised are "conservative" numbers baked in so that users don't push out towards the edges of the actual filter response.

3.  Related to question 2....the quoted BW for a given sample rate seems to be about 0.8 of the sample rate (even if they measure closer to 1 times the sample rate).  Certainly, the BW would have to be <=1...but was there any "scientific" reason for 0.8 specifically...or again, was it just a 'buffer' used to provide some margin for the actual sample rate?

Thanks!

---

Brandon

[Kyle_A.]

Hi Brandon,

1) I haven't yet found an explanation for this, but the equation for decimation defined as Decimation = 64M/Sample Rate, with coercion to one of the 8 discrete values referenced in the help for the 5660, this would lead me to believe there are benefits to having these decimation schemes when performing the calculations on the data.  While I cannot be sure, I believe it is related to the speed in which you can process these decimation rates.

2) We do spec with some room because you will see the bandwidth of your selected IQ rate, but the roll off for the filtering done here will not be instant but gradual. 

3) This spec gives the hardware the flattest response when it comes to acquiring data.  There are also a few filters on the device that are only flat through that particular BW (0.8 times the sample rate)

[Henrik_Volkers]

---

@cochenob wrote:

1. I'm sure there's an obvious reason for this....but why are the decimation factors limited to powers of 2?  i.e.  N=32, 64, 128, etc..

To calculate the mean value you add the values and have to divide by the number of items. Like division by 10 in decimal is easy (shift the decimal point) a division by 2 (or power of 2) in binary is done by shifting the value. Much faster than a calculated division 🙂