Amplitude in received signals in forward modeling with ultrasound signal (1MHz)

[shoaibanwar66]

Hi,

I am trying to compare SEM and FDM in USCT applications. I have been working with SPECFEM (a spectral element solver) for acoustic USCT applications and recently started working with deepwave. However, I noticed that the amplitude of the received signal in deepwave is way too low (in e-10). I am wondering if I am making any mistake. Would you please let know if there is any way to increase the signal amplitude in receiver end.

Here is the example code.

`import torch
import deepwave
import matplotlib.pyplot as plt
import numpy as np
import torchvision

plotting = True
v = 1500 * torch.ones(50, 50)

freq = 1e6
NT = 4500
DT = 2.5e-9

grid_spacing_orig = 2.5e-4
grid_spacing = grid_spacing_orig #4.1441559515042545E-005
dx = grid_spacing

cell_number = 50 / grid_spacing

#print (cell_number)

Src = [0.006, 0.010]
Rcv = [0.006, 0.001]

peak_time = 1/freq
source_amplitudes = deepwave.wavelets.ricker(freq, NT, DT, peak_time).reshape(1, 1, -1)
#print (source_amplitudes.squeeze().detach().numpy().shape)

source_locations=torch.tensor([[[Src[0]/grid_spacing_orig, Src[1]/grid_spacing_orig]]])
receiver_locations=torch.tensor([[[Rcv[0]/grid_spacing_orig, Rcv[1]/grid_spacing_orig]]])

print (Rcv[0]/grid_spacing_orig, Rcv[1]/grid_spacing_orig)

out = deepwave.scalar(
v, grid_spacing=grid_spacing, dt=DT,
source_amplitudes= source_amplitudes,
source_locations=torch.tensor([[[Src[0]/grid_spacing_orig, Src[1]/grid_spacing_orig]]]),
receiver_locations=torch.tensor([[[Rcv[0]/grid_spacing_orig, Rcv[1]/grid_spacing_orig]]]),
accuracy=8, pml_width=[2, 2, 2, 2], pml_freq=1
)

pred_signals_swap = out[-1]

if plotting == True:
y = pred_signals_swap.squeeze()
#print(y.shape)
freq = 1e6
nt=NT
dt=DT
plt.figure()
plt.plot(np.arange(0, dtnt, dt)-1.21e-6, source_amplitudes.squeeze().detach().numpy())
plt.savefig("STF_deepwave.png")
plt.figure()
plt.plot(y.detach().numpy())
plt.savefig("sig_pre_0_2")`

I have attached the STF and received signal images bellow.

[ar4]

Hi Shoaib, Thank you for testing Deepwave and for your message. I apologise for the delay in replying. Does dividing the output receiver data by dx**2 help? In Deepwave the sources will be the size of one grid cell, whereas the other codes you are comparing against probably use point sources, so the Deepwave amplitudes will be dx**2 times the size. Since you mentioned that the amplitudes are 1e-10 the value you expected, and your dx value is about 1e-5, I suspect that that is the cause. I hope that resolves your issue. Please let me know if it does not. Looking at the rest of your code, I see that you set the PML width to two grid cells. That is very small, and will result in the edges of your model being quite reflective. Is that intended? You also use a PML frequency of 1 Hz, but the actual frequency of your source seems to be 1e6 Hz. Did you mean to do that, or do you find the documentation unclear about how to set these values?

[shoaibanwar66]

Thanks. Dividing by dx**2 actually helped.
I need to go through the documentation about PML frequency again. No worries. Thanks again.