Weird sequence of USRP rx_timestamps when connecting a GPSDO.

Hello,

I am using an USRP B210+ GPSDO.

Recently, we can get timestamps of each sample from the USRP by using

[Samples,~,~,Timestamps] = Step(radio);

It performs well without a GPSDO. When setting the sample_rate as 20M(20e6) Hz ,the results may like this

According to the UHD WIKI, the timestamp of the first sample will be given in a rx_stream. Then Matlab automatically

calculate timestamps of each sample by Sample_Rate thereafter beautifully.

-----------------------------------------------------------------------------------------------------------------------------------

However, when plugging a GPSDO, timestamps in arrays show like this.

When I fprintf this Sample to see high decimal of them, by

fprintf("%3.30f",all_stamps(1,1) ); (2,1) (3,1) ......

I found something strange that, 4 or 5 sample are put the same timestamps.

The fprintf result looks like below.

1672916780.081696748733520507812500000000 Sample 3

1672916780.081696748733520507812500000000 4

....

1672916780.081696748733520507812500000000

1672916780.081696748733520507812500000000

1672916780.081696748733520507812500000000

--------------------------------------------------------------------- 与包头1差 2384 20M

1672916780.081696987152099609375000000000 8

1672916780.081696987152099609375000000000

1672916780.081696987152099609375000000000

1672916780.081696987152099609375000000000

-------------------------------------------------------------------- 与包头2差 2384

1672916780.081697225570678710937500000000 12

1672916780.081697225570678710937500000000

1672916780.081697225570678710937500000000

1672916780.081697225570678710937500000000

1672916780.081697225570678710937500000000

------------------------------------------------------------------ 与包头3差 2384

1672916780.081697463989257812500000000000 17

And everytime timestamps updates, the difference of the last timestamp is 0.000002384s,

which equals 5/(20*1024*1024).

-------------------------------------------------------------------------------------------------------------------------

To sum up, uncertain probability of occurrence 4/5 samples are put the same timestamps, and the timestamps

in interval equals 5 samples interval in 20Mhz SampleRate.

Howerver!! I look into my program that all I enter for SampleRate is 20e6, not 20*1024*1024.

Why did this happen?

Thanks.

[sy]

Why did this happen?

[天意]

The only one reason I can think that is GPSDO timestamps is more complex, and may use different data structures from the normal one.

To get a quick Timestamp Tagging of that complex data structure (double) ,the happening of weird timestamps may be due to using Shift Operation, since the interval is Power of 2.

However, this is just my guess. I hope to get other answers.

Answer 1

[天意]

It's because of the overflow of double64.

Answer 2

[Karunya Choppara]

This is an issue due to loss of precision, as the integer and fractional part are combined and assigned to double datatype output

This issue is resolved in R2026a, with a provision to choose the timestamp mode using TimestampMode property

By setting the TimestampMode to 'high-precision' we obtain the timestamp output as a struct, with fields 'timestampInt' and 'timestampFrac' representing the integer part in int64 datatype and fractional part in double datatype respectively.

see the properties and outputs for comm.SDRuReceiver system object