I tried to simulate a 4- channel WDM with QPSK on each channel. However, when the BER was too poor, I tested the channels individually . I was surprised to find that the BER is different among the individual channels. For the wavelengths of 193.5109 nm , 193.5359 nm , 193.5609 nm and 193.5859 nm, the BER for individual channels are 2.4e-2, 2.77 e-7, 2.5e-7 and 2.3e-2 respectively. The photodiodes and local oscillator are set to the source laser wavelength in each of the channels. I am attaching the .icp file qpsk_4ch_nofilt.icp (4.1 MB). I am unable to find the cause behind the BERs of two of the channels being different by 5 orders of magnitude when compared to the others.
If you remember from the previous post:
The BER is measured from the Q-factor by:
BER = 0.5*erfc(Q/sqrt(2))
and the Q-factor is calculated by:
Q factor = (Ptop -Pbase)/ (sigmatop + sigmabase)
For the 4 channels, if you look at the eye diagrams individually, you can find that none of them are evenly opened, but the channels at 193.5109 THz and 193.5859 THz have larger gaps between decision level 1 & 2 and 3 & 4 (which means they have a large sigma for 0 level & 1 level) while the channels at 193.5359 THz and 193.5609 THz have larger gap between decision level 2 & 3 (which means they have small sigma for 0 level and 1 level). That’s why the BER values are so far apart from each other.
In this case, I recommend to do post signal processing and calculate for the digital BER by comparing the transmitted and received signals manually.
I hope this answers your question. Let me now if you have further problems
Thanks for the suggestion. When I place the 193.5109 (193_5109.icp (1.1 MB)) channel in a seperate icp file and 193.5359 channel (193_5359.icp (1.1 MB)), the BERs are ideal (10e-307). But when I place them within the same file (2 chnl.icp (2.1 MB)), even though they are not connected to each other, there is a reduction in BER (10 e-6). My confusion is this interaction between the two schematics when placed in same icp file even though they are not connected to each other.
The problem you described indeed exists and we are working on to make INTERCONNECT work better on this aspect. The problem involves some complicated settings and calculations behind the sense, so it is hard to explain. But long story short, INTERCONNECT automatically estimate the bandwidth required for multi-channel systems. It creates a new simulation bandwidth that contains the individual signal channels. In this process the center frequency of the simulation bandwidth is defined automatically. This may be an issue with the waveguide/modulator models. The waveguide model calculate the group delay and phase shift based on the center frequency of the simulation bandwidth, the issue is that as you add more channels the values taken will change since the center of the bandwidth is changing. This is why adding more channels may lead to inconsistencies when running time domain simulations.
To solve this problem, there is a property “Sample Mode” in the “Simulation” property settings, you can enable the “sample mode frequency band” by setting it to “single” and type in the “sample mode center frequency” you want. This will lock the center frequency and bandwidth, so the simulation will give same results for individual channels no matter how many new channels you introduce to the system. But depending on the center frequency you choose, the simulation results are different. We are working on a better solution to this problem and to make the system modeling more realistic.
To follow up our discussion for the BER issue, the ‘Eye Diagram’ element can actually take in the bit pattern input. Then if you select the ‘BER estimation’ to be ‘measured’, the ‘Eye Diagram’ will count for the error bits accordingly. However, it is better to do a normalization before this step since sometimes when the signal level is too low, they all will be treated as 0s.
I hope this could help. I will definitely keep you in the loop for our progress on this issue.