Thank you for your commends. After I delete the two lines below:

1. include <stdio.h>
2. include <iostream>

The Flash and Ram are both smaller: Flash: 35.1 KB, Ram: 0.9 KB. Now sample_numb = 2048 is working. But for bigger sample_numb, 3072, 4096, still not working.

I tried to paste the program here, but it was limited for the size 3000, how can I do that? Thank you.

The link is below:

https://developer.mbed.org/users/ZZhang/code/AC_sampling_05_1768_discussion/

For something that size the best option is to publish it and post a link here. In the compiler click on commit, that saves your current code into the version control system. Then click on the down arrow next to commit and click publish. Select to publish it as a program and make it either public (will show up in search results) or unlisted (anyone can view it but it won't show up in search results). You can then post a link to the project page here and other people can view or import the code.

At any time you can go to the project page and under Admin settings you can change the project to private so that no one else can see it.

Any changes you make after hitting publish are private to your account unless you commit it and then publish it again. When you publish a version all of the committed versions prior to that point are included in the project history.

Thank you, Andy. The link of the code is below, please have a look. Thank you.

https://developer.mbed.org/users/ZZhang/code/AC_sampling_05_1768_discussion/

As expected almost all your memory is being allocated dynamically. Around line 53 of AC_sampling_05.cpp you have

    sample_valu = new float [sample_numb];
    period_record = new int [sample_numb];  
    max_array = new float [sample_numb];

The new command dynamically allocates memory for the data while the program is running. The compiler can only report memory that is allocated before the program starts to run. You are requesting 2 floats and 1 int = 12 bytes per sample. 2048 samples = 24k. Further down you use new to create another array of 101 floats, that's another 0.4k.

The compiler indicated 0.9k used so 25.5k are accounted for.

Almost all of your objects are being created on the stack (e.g. your Samples_stru object is local to main rather than being global) and so their memory requirements also aren't being reported by the compiler.

In short after the first 10 lines of main() your memory usage has gone from 0.9k to around 30k.

If the system can't allocate enough memory when you call new it returns a null pointer. You never check for this possibility, your program tries to use the invalid pointer and crashes. For debugging the following changes

    sample_valu = new float [sample_numb];
    if (!sample_valu)
      pc.printf("OUT OF MEMORY allocating sample_valu!\n");
    period_record = new int [sample_numb];  
    if (!period_record )
      pc.printf("OUT OF MEMORY allocating period_record !\n");
    max_array = new float [sample_numb];
    if (!max_array )
      pc.printf("OUT OF MEMORY allocating max_array!\n");

and similar lines after every instance of new or malloc would make this sort of thing a lot easier to find in the future.

As a quick fix put these two lines right at the start of AC_sampling_05.cpp

static float sampleBuffer[4096] __attribute__((section("AHBSRAM0"))); 
static float maxBuffer[4096] __attribute__((section("AHBSRAM1"))); 

and then in void Samples_stru::Initialization() change the values of sample_valu and max_array to being

    sample_valu = sampleBuffer;
    period_record = new int [sample_numb];  
    max_array = maxBuffer;

That will move the two float buffers out of main memory and so free up a lot of memory without requiring any significant code changes. Everything should then work for values of sample_numb up to 4096. Go over that and everything will fall to pieces and crash.

Going over that size would need some significant reworking to reduce memory usage.

As a general style point dynamic memory allocation like this isn't a good idea for embedded code, it has it's uses but should be avoided whenever possible. If you know how much memory you will need when you compile the code then allocate it statically instead if using new or malloc, it's far easier to debug code when the compiler can tell you the problem rather than trying to track down the cause of a random crash.

Thank you very much Andy! Now for sample_numb = 4096 is working.

It seems for LPC1768, the case of sample_numb=8192 is impossible, is it true ?

If it was that, for QSB LPC4088, it should be no problem? Thank you.

Your memory useage is about 8k + 12k for each 1024 samples. So 2048 samples takes 32k, 4096 takes 58k and 8192 samples would take 104k.

So any mbed with over 104k of ram should work for 8k samples.

It may also be possible to re-work the code to reduce the memory usage.