Troubles at beginning of blocks in BlochDict simulations

[Tooine]

What happened?

Hi,

I'm using BlochDict to create dictionaries of signals for MRF sequences.

Signals look fine when I look at the magnitude. Yet, checking the real and imaginary parts, I observe unwanted peaks in the signal evolutions.

Attached are some figures showing the desired signal (1 block) and the signal obtained with 10, 20, 100 and 130 blocks.
The peaks are regularly spaced along the sequence and the number of them seems to be the number of blocks sim_params["Nblocks"], at least when this number divides the number of pulses of the sequence (260), which indicates that the bug might be depending on the type of the first event of the block (for 100 blocks, only a few peaks are observed ...).

Is it a known bug or do you have any guess what could be the problem? I'm also attaching the notebook and pulseq sequence if necessary.

Edit: this bug does not appear when I use Bloch instead of BlochDict with a single spin. It appears both with GPU and CPU simulations. Also, the peaks appear for spins with a (large) non-zero Δw, this might gives an idea of the problem?

Koma_bug_Nblocks.zip

Environment

OS x86_64-linux-gnu
Julia 1.11.5
KomaMRIPlots 0.9.1
KomaMRIFiles 0.9.1
KomaMRI 0.9.1
KomaMRICore 0.9.1
KomaMRIBase 0.9.1

[cncastillo]

Thanks for the fantastic bug report, this is definitely a high-priority problem. The relation with off-resonance is a good pointer to start looking into this.

@Stockless After fixing this, a test to ensure that this does not happen again will need to be made.