What is Channel Modeling

In communications a Channel is the medium used to transmit information from one point to the other. Ideally, modeling a channel takes into account all the physical effects on the signal as it propagates from the transmitter to the receiver.

I am using HFSS TDR and compare with Agilent TDR in measurement of a sliced via, I found usually HFSS has 2 or 3 ohm higher than Agilent, why?

It is very instructional to play with the wide variety of windowing functions to deal with the iFFT on band limited frequency domain data and see how it affects the rising and falling edges which in turn has a direct impact on the peaks and valleys of impedance discontinuities that are shorter than the rise time length. Starting with simple test structures like a series resonant Beatty standard (ie 50ohms to 25ohms to 50ohms) and varying the length of the discontinuity can be very instructional when looking at time and frequency domain and matching measurement with simulation.

Some interesting observations to explore:

1) HFSS is an FEM simulator which is a frequency domain solution, so it is similar to a network analyzer measurement with extrapolation to DC and band limited with a max frequency. Ideally this max frequency should be higher than the bandwidth of the DUT, but for simple passive structures this is not always easy to do. In simulation increasing the max frequency increases the mesh and the simulation time, in measurement not only does the instrument have a max frequency, but the ability to design a wide bandwidth fixture and calibrate it out also has its bandlimiting max frequency. As the max frequency gets lower and closer to the DUT frequencies of interest, then one also becomes more sensitive to the “windowing” that is being used for the conversion to the time domain.

2) On the measurement side with a TDR/TDT instrument one now has a broad band step edge which can be calibrated to correct for fixture losses, but identifying an industry accepted standard for the shape of that rising and falling edge is not a simple thing. So in short if you send an “unknown” step edge shape into a black box then it will be difficult to match with simulation. Also, the TDR/TDT measurement does not require an FFT transformation which is typically non-causal due to the required windowing.

3) Matching the TDR from an iFFT of a frequency domain simulation with the TDR from an iFFT of a frequency domain measurement is much easier to do since one can insure that the same windowing is being done and identical reference planes are being used.

4) To get simulation to match time domain TDR/TDT measurements can be done with a few different approaches. One is to use a time domain EM simulator (Finite Difference Time Domain FDTD) and stimulate with a stored TDR step edge waveform that was obtained from measurement of the output from the TDR box at the reference plane of interest. Second is to use this stored waveform and run it through the frequency domain simulation data using a causal Hilbert transform instead of the iFFT that requires non-causal windowing techniques.

As with anything it always helps to start with something simple that is easy to understand. Al Neves of Wild River Technolgy has some great connectorized test structure boards to 50GHz, and I am working with him to do an ADS simulation starter kit to demonstrate these issues and how to get simulations and measurements to match. One of my favorite structures is the simple series resonant Beatty structure (ie 50 ohms to 25 ohms to 50 ohms) that works great for validating both simulations and measurements in the time and frequency domains.

If you can’t measure and simulate a simple series resonant test structure correctly, then how do you expect to do a complicated SERDES channel with packages, PCBs, connectors, and cables?

… and in response to the differences you are seeing #1 is that the fixture is slowing down the risetime so that the Z peak is lower for a via discontinuity that is shorter than the rising edge and you need to do fixture removal calibration techniques or, #2 the HFSS max frequency is allowing higher amplitudes at the higher frequencies then the high frequency roll-off that is typical in a TDR step edge, or #3 the as-fabricated material properties are not the same as used in simulation (ie drill size vs finished hole size, etching, lamination thicknesses, etc).

What is the Channel Modeling Program?

The Channel Modeling product is a really a program where we invest time with the customer over a 3 month timeframe (based on your needs and schedule), we work with you to analyze S-parameter quality, correspondence approach, review your goals, help with any presentations, and importantly assist with developing a causal/passive de-embedding or full path calibration approach.    You can compare our measured parameters with your parameters are simply use the supplied S-parameters to get a quick start for the correspondence effort.    Our customers have seen a real benefit of the entire package.   The program assistance is included in the cost.

Finally, we offer comprehensive VNA training, including on site 1.5 days hands on classes.    WRT can share our VNA set up tips, calibration verification, and S-parameter quality work flow.

Almost all of the  EDA companies and all of the major capitol equipment companies are our customer, including CST.