convert Simulink float-point data to FPGA fixed-point data

[GAOY0806]

I built a model in simulink, whose data type is double. Then I convert it's data type to fixed-point and the altered model ran successfully in simulink.

But when I converted the fixed-point simulink model to VHDL file and put the file into FPGA(through CLIP node), the simulation results are wrong.

I guess I made some mistakes in data type conversion(float to fixed) which cause data loss but have no specific idea.I paste a screeshot and wonder if there is something wrong. (I can make sure that the set fixed-point type prevent overflow happening but not sure if there is a mistake in operation like plus and times).

[RavensFan]

Did you post the correct attachment?  The diagram you attached doesn't show anything about a fixed point/floating point conversion.

[GAOY0806]

The attchment shows the converted data, which means conversion of float-point to fixed-point is completed. For example, sfix(32,31) represents 32 bit word length and 31 bit decimal length, which corresponds to FXP(+/-,32,1) in Labview FPGA.

For example, 2*2=4. I set them as (+/-,16,4), (+/-,16,4) and (+/-,16,5). I think (+/-,16,5) covers 4. But as I know, the result data type should be set as (+/-,32,8). I don't know if (+/-,16,5) is feasible.

In another simple case, I converted int32 to (+/-,16,10), which covers the full range of the data. The model runs correctly in simulink but wrong in FPGA. When I changed (+/-,16,10) to (+/-,32,26), the model runs correctly both in simulink and FPGA.

So I Think maybe there is some mistakes in my FXP configuration, which although covers the full range of results, but still wrong maybe due to integer/decimal word length settings. For the simple case, I find the problem luckily ,but for this complicated model, I can't identify the problem.

So I wonder if there are some general rules in converting float-point to fixed-point and word length configuration rules regarding fixed-point operation.

[RavensFan]

No, your attachment doesn't show any converted data.  It shows some sort of functional flow chart.

I realize your subject line talks about Simulink.  That isn't an NI product.  Perhaps you should post your question in a forum dedicated to that software.  You talk about LabVIEW FPGA in your message, but you haven't shown anything about LabVIEW in your attachment.

[GAOY0806]

Hello RavensFan ,

This attachment shows how FPGA CLIP communicates with Veristand in our project. The conversion of data type is completed in Simulink, which is also proved correct in Simulink simulation. The conversion work done in Simulink is just a preparation for the transfer of simuink model to FPGA CLIP, which needs fixed-point rather than float-point. But I think the conversion I have done have some mistakes which can't be presented in Simulink.

The conversion problem is just my guess, because the model runs successfully in Simulink but wrong in PXI. 

Thanks.