Actions
Forum Thread Details
  • State Suggested Answer
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 473 subscribers
  • Views 684 views
  • Users 0 members are here
Related

Storing Data Points in Embedded Systems Efficiently and Reliably

kas.lewis

I am looking for an efficient way to store data points in an embedded system. More precisely, as part of the Sudden Impact Challenge, there is a possible need to sample the heart rhythm and store that data for analysis. The simplest method I can think of would be to make a very large 2D array to put those samples in but doing this sampling at 500Sa/s (12 bit ADC) would consume any available memory after just a few samples.

 

If there are any suggestions or proven techniques I would be grateful to hear them.

 

Thanks

Kas

Parents
  • 0

    Hi Kas,

     

    One option could be to store 1-in-100 full waveforms. For example, if the heart is beating at 100bpm, then perhaps just one fully captured beat (at the 500 samples/sec you mention) could be sufficient to identify anomalies within a minute. For the rest of the time, just basic information such as pulse rate could be captured. This would reduce the storage and data transmission requirements to 1/100th.

    Another option could be to store deep data (e.g. on a multi-GB microSD card), but only serve up data once per (say) minute, and allow the remote system or operator to selectively download more data as needed, dynamically.

    There are probably smart compression schemes too but could be complicated math and need a lot of local processing.

    Hard to say what scheme is best without knowing from a medical expert I guess! For example perhaps 1 minute may be too long, but a medical expert may think 10 seconds is fine.

     

    But Clem's idea of local processing is smart. Nowadays there is maybe enough low-cost (and low power) computing resources to do this.

Reply
  • 0

    Hi Kas,

     

    One option could be to store 1-in-100 full waveforms. For example, if the heart is beating at 100bpm, then perhaps just one fully captured beat (at the 500 samples/sec you mention) could be sufficient to identify anomalies within a minute. For the rest of the time, just basic information such as pulse rate could be captured. This would reduce the storage and data transmission requirements to 1/100th.

    Another option could be to store deep data (e.g. on a multi-GB microSD card), but only serve up data once per (say) minute, and allow the remote system or operator to selectively download more data as needed, dynamically.

    There are probably smart compression schemes too but could be complicated math and need a lot of local processing.

    Hard to say what scheme is best without knowing from a medical expert I guess! For example perhaps 1 minute may be too long, but a medical expert may think 10 seconds is fine.

     

    But Clem's idea of local processing is smart. Nowadays there is maybe enough low-cost (and low power) computing resources to do this.

Children
No Data