I used the PSO algorithm to find the optimal solution for MCMA in this algorithm, and the generated constellation diagram showed that the modulus could be recovered, but the error rate was high. I found that the phase of the output signal has deviated by π n/2. For example, after the signal in the first quadrant is balanced, it changes to the fourth or second quadrant. Through analyzing the code, I found that the initial particle position may be random, which caused the generated filter to have a significant impact on the input signal, resulting in confusion. How can I modify the initialization of this particle position to correctly restore the original signal?