DeboraHarry
Full Member level 5
I'm trying to simulate a small antenna (100 mm long, 10 mm diameter, some parts less than 1 mm in size) over a large frequency range (200 MHz to 5 GHz) in HFSS.
It it taking ages. It has been running 3 hours, has used 17 GB RAM, and is still on only the 2nd adaptive pass. At this rate I can imagine it will take several days at least. (The machine has far more than 17 GB RAM, so its not swapping to disk). Despite my other questions about setting up for best use of parallel processors, it is clear this part of the simulation is using all 8 cores, as the load average is close to 8. Here's the output from "top" which Linux users will know. For Windows users, consider the task manager would show the CPU usage is very high.
I think the problem is this, but I'd appreciate a second opinion, and any advice.
1) To get the radiation pattern one needs to use a small mesh on the radiation box. That need to be lambda/6 or smaller at the highest frequency, so length of a tetrahedral on the radiation boundary needs to be 10 mm or less for accuracy at 5 GHz.
2) To get accurate results at the low-frequency, (200 MHz), the radiation boundary needs to be a distance of at least 750 mm (lambda/4) away from the antenna
So I end up with a large volume of airbox to ensure accuracy at the low frequency, and a fine mesh on the radiation boundary to ensure accuracy at the high frequency. But neither of these are simultaneously necessary. The airbox needs to be large for the low frequency, but the mesh does not need to be very fine on the radiation boundary. At the high frequency, the mesh needs to be fine, but the airbox does not need to be large.
So I'm thinking I should split this problem into a 2 (or more) different simulations, rather than 1. Perhaps if I split it into 2, use one to cover the range 200 MHz to 1 GHz, and a second to cover the range 1 GHz to 5 GHz. Then using a small mesh and small airbox for the 1-5 GHz simulation, and a larger airbox and larger mesh for the 200 MHz to 1 GHz simulation.
Am I missing a trick here? Is this a sensible approach?
Deborah.
It it taking ages. It has been running 3 hours, has used 17 GB RAM, and is still on only the 2nd adaptive pass. At this rate I can imagine it will take several days at least. (The machine has far more than 17 GB RAM, so its not swapping to disk). Despite my other questions about setting up for best use of parallel processors, it is clear this part of the simulation is using all 8 cores, as the load average is close to 8. Here's the output from "top" which Linux users will know. For Windows users, consider the task manager would show the CPU usage is very high.
Code:
top - 17:15:18 up 22:07, 3 users, load average: 7.63, 5.46, 4.37
Tasks: 209 total, 1 running, 208 sleeping, 0 stopped, 0 zombie
Cpu(s): 81.8%us, 0.7%sy, 0.0%ni, 17.5%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 44927468k total, 17408008k used, 27519460k free, 157816k buffers
Swap: 35291128k total, 0k used, 35291128k free, 5860432k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
17544 deborah 17 0 17.7g 10g 15m S 659.9 24.5 41:56.29 hf3d
10630 deborah 16 0 12764 1184 820 S 0.7 0.0 1:46.66 top
1 root 15 0 10372 696 584 S 0.0 0.0 0:01.91 init
I think the problem is this, but I'd appreciate a second opinion, and any advice.
1) To get the radiation pattern one needs to use a small mesh on the radiation box. That need to be lambda/6 or smaller at the highest frequency, so length of a tetrahedral on the radiation boundary needs to be 10 mm or less for accuracy at 5 GHz.
2) To get accurate results at the low-frequency, (200 MHz), the radiation boundary needs to be a distance of at least 750 mm (lambda/4) away from the antenna
So I end up with a large volume of airbox to ensure accuracy at the low frequency, and a fine mesh on the radiation boundary to ensure accuracy at the high frequency. But neither of these are simultaneously necessary. The airbox needs to be large for the low frequency, but the mesh does not need to be very fine on the radiation boundary. At the high frequency, the mesh needs to be fine, but the airbox does not need to be large.
So I'm thinking I should split this problem into a 2 (or more) different simulations, rather than 1. Perhaps if I split it into 2, use one to cover the range 200 MHz to 1 GHz, and a second to cover the range 1 GHz to 5 GHz. Then using a small mesh and small airbox for the 1-5 GHz simulation, and a larger airbox and larger mesh for the 200 MHz to 1 GHz simulation.
Am I missing a trick here? Is this a sensible approach?
Deborah.