Failed to allocate TF Arena error but RAM has space

Help
#1

Hello i’m having this TF Arena error which i know from the forum is related to memory issue…
However i’m confused… the estimated peak RAM Usage is 104.4 KBs reported by edge impulse and i’m using nano 33 ble sense which has 256 KBs of RAM…

What could be the issue… the same model works for 1 sec of window size but doesn’t work for 2 second of window size. Please guide… it’s urgent

#2

Hello @rida,

Can you share your project ID please?

Best,

Louis

ERR: MFCC failed (-1002)
#3

Hello @louis

156911 is my Project ID. The TF arena issue is solved through boards.local.txt thing… but now it complains about mfcc!
Since i want to test this model… i don’t prefer to change the settings but any correction is welcome!

#4

Hello @rida,

Are you using the continuous inferencing sketch example or the normal one?
I can see that the inference time from the studio when I selected the Arduino Nano 33 BLE Sense is

  • DSP: 1159ms
  • NN: 434ms
    If you’re using the continuous mode, the impulse several times per window depending on the value you set here:
    #define EI_CLASSIFIER_SLICES_PER_MODEL_WINDOW 3
    If your window size is 2000ms and you want to run 3 inferences during that window but each takes about 1500ms to run, you will see in a memory error.

I hope that helps

Best,

Louis

#5

Hey… no i’m not using continuous example! i’m using the normal one…

#6

I have tried again with the non continuous example… it gives the same error about mfcc!

#7

Also… in this project i can not build the quantized version. It gives error and i have to go for unoptimized…

#8

Update… when i downloaded the quantized version of library by turning off the eon compiler (since it wasn’t working). The error i’m getting now is this… with same example

linker_script.ld:138 cannot move location counter backwards (from 000000002010c3c0 to 000000002003fc00)
collect2: error: ld returned 1 exit status
exit status 1
Error compiling for board Arduino Nano 33 BLE