SNR

Hi,

I have to admit that in reading all of the literature available from Nortek, I can not figure out the SNR values. I took data in a particle rich environment and recorded good results and the SNR levels ranged from about 5-20 units. I took data elsewhere where we suspect that the number of particles are much fewer and after removing all of the data with correlation less than 70%, I get SNR that is in the range of 6-20. I would like to use the SNR to clean additional noise from the data after removing all data triplets where any of the three components had correlation less than 70.

I do not understand how SNR is reported. I will give my view here and hope that someone will set me straight. The first sample cycle of a run, there is no ping and the Vector collects a background noise level. Hopefully this is true for power up after power was off as well as data taken in burst mode. My understanding is that this is only collected once per burst or power up. Then data collection starts and the strenght of the returns are measured as the parameter amplitude. Then it would make sense that the SNR would be the amplitude over the first ping (as recorded in the vhd file) but the units then are counts/counts and hence non-dimensional. However the SNR reported in the data files has units of m/s as told in the hdr file and in the manual. So I am confused what SNR is and how I can use it... what is a threshold for bad data...

thanks in advance,

Dave

Ok, re-reading the discussion on de-spiking illustrates that the SNR comes from the subtraction of the noise from the signal amplitude. Then the conversion from counts to dB is applied. I guess then my question is how do you convert from dB to m/s and what is a good cutoff value or how do I interpret the SNR with respect to when the data are good and when they are suspect (in the case of suspecting low particle densities)?

Cheers,

Dave

Hi Dave,

The unit for the SNR is dB, there is a typing error in the manual. I checked the hdr file as well and found that the hdr file correctly reports dB as unit for SNR.

You have a good understanding of how the SNR is calculated, the amplitude levels measured in counts during velocity measurement are subtracted with the noise level in counts. The noise level is measured at the beginning of each burst (and at the beginning of continuous measurements). This non-dimensional SNR value in count is then multiplied by the scaling factor of 0.43 dB/count, to arrive at the SNR in dB.

As for removing noise from your data set the best way is to look at the correlation like you have already done. The correlation will precisely tell you the quality of the data.

Note that in the burst header file (vhd) the correlation of the noise is also output, typical values are below 10% which is expected since noise is uncorrelated. As the SNR decreases the velocity correlation will approach this value. So when you discard the data with a correlation below 70%, you are screening the data with respect to the SNR as well. " />'> This is why you end up with the same SNR numbers after discarding samples with correlation below 70%.

The best time to look at the SNR is if your data have correlation values close to 100%, because you then no longer have information about the amplitude in the correlation value. This will be at SNR values above 15 dB. When you have high SNR values, you can choose to reduce the output power level from HIGH to LOW in the advanced menu in deployment planning. The reduction in SNR will be approximately 7 dB. This means that if you see SNR values above 22 dB throughout your data set, you can choose to reduce the output power to extend the deployment time.

Cheers,

Sven

Thanks Sven,

One quick followup. We are not using the vector in a conventional way, by using a burst mode. We are controling the collection of data by turning the power off and on to the vector. It was originally set to continuous collection mode and we have to somehow attach a binary hdr file for each time we turn the instrument on. The data is collected externally on a flashcard on the platform to which the vector is attached. In this way, does it still record the noise level each time that the vector is given power? Also can you say how much delay there would be from when the vector receives  power to when it starts to take data? I.e. is there some startup or warm up time? It seems that the platform has done some things before I start to receive data from the instrument. I was not involved in the interfacing or setup of the instrument, I am the recipient and user of the data. If you have further questions for me, I can find them out.

Thanks in advance,

Dave