Simulation of RF transmitter in SpectreRF

[KostasRF]

I am trying to simulate my transmitter (TX) design in SpectreRF (Cadence). I have created a schematic with ideal components (from analogLib) and Verilog-A models for the digital gates. I want to simulate the complete TX using a real cellular signal, such as CDMA or LTE. For this purpose, I have created these signals in MATLAB and I am using co-simulation between Simulink and SpectreRF. The data from Spectre is fed to Simulink where I do all post-processing for spectrum analysis.

My TX uses a switched capacitor technique with a high frequency clock, so I am forced to use a very small time step in the transient analysis, which results in extremely long simulation times. To get the resolution bandwidth I need, I have to run the transient analysis for at least 100us with a time step that is in the order of ps! I have tried envelope analysis, but it doesn't work nicely with co-simulation (doesn't capture the switched capacitor behavior). Any ideas how I can speed up the simulation? I want to be able to simulate EVM, ACPR, and PSD.

Thanks in advance!

 

[BigBoss]

Since you're using ideal elements, why don't you try this simulation in Keysight's ADS if it's available ?? Cadence Spectre is not a proper simulator for those type of problems.
But you're trying do a really hard task.My suggestion is to use parallel simulators if there are other machines.You IT guys may help you to install LSF.

 

[KostasRF]

Since you're using ideal elements, why don't you try this simulation in Keysight's ADS if it's available ?? Cadence Spectre is not a proper simulator for those type of problems.
But you're trying do a really hard task.My suggestion is to use parallel simulators if there are other machines.You IT guys may help you to install LSF.

Thank you very much, BigBoss. I have access to ADS, but I'm not sure how I can make it work with Simulink. Does ADS support PWL sources like SpectreRF, where I can load the baseband CDMA samples?

 

[BigBoss]

Thank you very much, BigBoss. I have access to ADS, but I'm not sure how I can make it work with Simulink. Does ADS support PWL sources like SpectreRF, where I can load the baseband CDMA samples?

ADS Ptolemy Simulator can work with MatLab.And there are many complex architectural functional blocks in ADS.I don't know there is PWL for Ptolemy but I'm sure there are other similar blocks.
You can also -perhaps-find a "ready-to-go" set-up in ADS for your purpose.

 

[pancho_hideboo]

Cosimulation between Spectre and Simulink is very slow.

https://www.edaboard.com/showthread.php?t=206326#2
https://www.edaboard.com/showthread.php?t=364511#4

ADS Ptolemy Simulator itself is far slow.
Cosimulation between ADS AnalogDomain(Tran or Envlp) and Ptolemy is very very slow.
Keysight will not improve it any more.
Ptolemy is almost dead.
Current SystemVue Electronic System-Level (ESL) Design Software in Keysight is SystemVue not Ptolemy.

However I recommend to use cosimulation between Spectre and NCSim(Incisive) which is also sometimes called as AMS-Designer.
Here use SystemVerilog with DPI-C.
https://mathworks.com/help/hdlverifier/ug/dpi-c-generation-overview.html
https://mathworks.com/help/hdlverifier/ug/generate-cross-platform-dpic-components.html
https://www.mathworks.com/videos/generating-dpi-c-models-from-matlab-using-hdl-verifier-100430.html

 

[KostasRF]

Thank you, pancho_hideboo! So, if I understand correctly, you're suggesting I generate a SystemVerilog component using the HDL Verifier in Simulink and then using this as a behavioral model in AMS-Designer. Would this, however, solve the issue of sharp transitions in my high-speed square-wave clocks which tend to slow down my PSS and transient simulations in Spectre?

 

[pancho_hideboo]

So, if I understand correctly,
you're suggesting I generate a SystemVerilog component using the HDL Verifier in Simulink and then using this as a behavioral model in AMS-Designer.

Little different.
I recommend you to use DPI-C model via SystemVerilog.

Would this, however, solve the issue of sharp transitions in my high-speed square-wave clocks which tend to slow down my PSS and transient simulations in Spectre?

Yes.

First, can you understand PSS correctly ?
Considering PSS analysis is nonsense in System Simulation.

Meaningful simulations are Transient and Envelope Analyses.

Second, can you understand advantage of NC-Sim ?

NC-Sim is an event-driven type simulator for signal-flow model.
Simulink is a fixed-time step or adaptive-time step simulator for signal-flow model.
Spectre and Spice are an adaptive-time step simulator for energy-conservative model.

https://www.designers-guide.org/Forum/YaBB.pl?num=1238242506/7#7

So NC-Sim is most effective for signal-flow model.
See "12.4 Efficiency of the Models" at page-39 of https://www.designers-guide.org/Analysis/PLLnoise+jitter.pdf

 

[KostasRF]

Thank you again! I am not familiar with System Simulation, as this is my first time designing a complete TX. Your feedback has been very valuable!