为什么mode source 的发射持续时间的长短会比较显著的影响频谱


#1

我的模型中,mode source 发射持续时间的显著改变会严重影响我的频谱,这是为什么呢?我的模型如下


How to plot the exactly transmission
#2

Hi @phd.wangbo,

你好。可以上传你的仿真文件吗?方便我们找出问题。:slight_smile:


#3

Hi @gwang
你好!我的仿真文件是实现这样的功能,发射光源w1经过折射率的调制区域(调制函数为ε(t)=ε(max)cos⁡(ωt))后,只有频率为w2通过
我对仿真的文件有两个问题:
1,随着时间的变化,我的频谱图也会相应发生变化。例如,当我在time domain setting那里设置pluselength=3000fs,offset=6000fs时,光源发射时间大概为0.15e5 fs,此时的频谱图为


当设置pluselength=20000fs,offset=40000fs时,光源发射时间大概为1e5 fs,此时的频谱图为
继续设置其他发射时间,频谱图会继续变化,请问怎么解决?
2,后来想用透射率图来代替频谱图,可是,发现plot出的图形不是很可靠,在于没有归一化。
现附上我的仿真文件,和相应的pulgin material的插件源代码(timespatial.h).txt (1.2 KB)
(timespatial.cpp).txt (1.4 KB)
Complete optical isolation.fsp (281.7 KB)
期待你的解答。


#4

mode source 发射持续时间影响频谱宽度以及光谱分布;另外材料特性随时间变化,因此改变时间参数结果可能会变。
你可以将结果除以光源功率


#5

Hi @gsun
谢谢老师的回答。我还有几个问题想咨询老师,
第一个问题是,在我的仿真模型中(从楼一可以看到),在调制区域里,我添加了mesh override region,选择equivalent index x=20 and y =20,得到的频谱为下面(是可以接受的结果)


接着我想让绿色的频率更小一点,所以提高equivalent index x=50 and y =50,得到的结果却比较诧异(结果不能接受了)
然后,我再试一下继续提高equivalent index x=60 and y =60,得到的频谱图又出现比较满意的结果
请问,为什么提高equivalent index 会出现这样的结果?怎么样的情况才是可信的?

第二个问题是,继承问题一,问局部精度是否越大越好(省略考虑模拟时间的情况)

第三个问题是,网格精度的增大,这个autolevel需要进一步降低吗?

第四个问题是,你提到的“另外材料特性岁时间变化,因此改变时间参数结果可能会变。”,我的这个模型是仿真nature photnoic 的一篇文章,按照这篇文章的报道,结果应该就是w1作为input , 只有w2作为output。附上参考文献为here

期待老师的解答!:slight_smile:


#6

A1:用时间监视器能得出谐振频率,但是谐振的强度就与时间监视器的位置和网格精度有关,另外也跟仿真的时间长短有关(你看看分析组是如何取时间窗口的)。有5跳到50步幅太大了,如果你想测试,需要慢慢增加,此外还要考虑PML的总厚度最好爆出不变;
A2:局部精度同时影响其它地方的网格精度,但PML不在均匀网格区域时性能变差。一般没有必要用equivalent index x〉10的,除非你有特殊原因,同时你还要看是不是小于 min mesh step(缺省为0.25纳米);
A3:一般没有必要,除非谐振特别强或者你发现了相关问题;
A4:抱歉我没法阅读连接的文献(没有订阅)。当折射率随时间变化时,输出也可能随时间变化。请详细阅读文献,看看它的输出结果是稳态的还是瞬态的?如果是瞬态的,从那个时间段计算?

另外,一个帖子的内容应该与题目一致,当问题改变后请另外发帖。
我分别用equivalent index=5和10测试,结果没有什么区别:


但是Decay曲线对应的蓝色结果并不是直线因此计算的Q不准确。

需要注意的是,计算Q和谐振波长域计算透射率反射率所需要的仿真时间不一样,前者很短时间就可以,不需要等到信号彻底衰减,否则截取的时间信号可能太宽而不能精确计算;而后者必须用很长的仿真时间使得信号彻底衰减得到精确的频域结果。


#7

谢谢老师的详细解答!


#8

the transmission already divides the output power on the input power. I didn’t get the sentence “You can divide the result by the light source power”.


#9

this is because we are talking the spectrum directly from time monitor, not from frequency-domain monitor.


#10

I’m getting a transmission in the range of e+12. This is completely not making any sense. I used the same time limits like in the design posted. However, I didn’t use an override mesh, instead, I used a mech accuracy in the FDTD region of 8. The answer shall be correct, right?

Simulation time=1e+6 fs, pulse width=20000 fs, offset=40000 fs. what shall I do?


#11

I guess your result is from the spectrum of the time monitor. if so, please try to normalize your result with spectrum of the source signal using the same FFT method.


#12

No, it’s from the power monitor transmission not the time monitor.

I’m getting this shape from the power monitor.

Another question, is using the time monitor an accurate method to obtain the transmission? It only calculates the power at 1 point not the whole cross-section, right?


#13

without checking your simulation file, it is hard to say. If you have a gain material, the transmission can be quite large; and if the frequency is far away from the source spectrum, its source power can be very small so large value divided by small value can be a problem.

only when your device is periodic and the transmission is a plane wave then you can use a point time monitor to measure it.


#14

I’m doing the same idea and same design posted above. I guess I got your point in the very large transmission value. I’m inserting a wavelength of 1.55 um and waiting for output power at 1.12 um so the transmission might be very large for this. I have 2 questions:
1- Shall I ignore the large transmission value and say that since power is interchanged between 1.55 um and 1.12 um then the design is working properly?
2- Is using the time monitor is right in this design since neither the structure is periodic nor the excitation is a plane wave. If yes, I didn’t get your point in normalizing it, how shall I do this?

Thanks.


#15

A1: you may need to understand the cause. If it is really due to “nromalization” you can say so. However if your simulation is diverging or some other reasons, you need to dig it out.
A2: if it is not a plane wave, you will need to use frequency-domain power monitor. It integrates the power over the transmission plane.


#16

It’s not giving me any divergence errors and also the mesh settings shall not result in stability problems as I’m using auto non-uniform mesh. Thanks a lot for your help.


#17

Good luck!