RF Simulation Troubles in PSPICE

You can, but the accuracy you get is probably not going to be that great -- especially for something like a mixer, which is necessarily a non-linear device, so coming up with an accurate SPICE model is very difficult unless you settle for linearizing ing it about your particular operating point. What one usually does if they need a highly accurate model of a part that doesn't come with such a model to begin with is to obtain a sample of the device in question and sit down and characterize it themselves -- this is a time-consuming process, and requires a fair degree of experience to do well (i.e., unless you can find someone in your department who's done it already and can help, I wouldn't suggest doing it unless this is strictly a "learning experience" kind of project).

What source of device are you building, and what parts of it do you want to simulate? Although you can use SPICE for RF simulation, the results are very slow to achieve since transient analysis is usually the only kind of SPICE simulation that'll work OK for mixers and modulators, and the time steps get to be painfully small if you're dealing with high frequencies (as it would appear that you are form your parts specs).

I'm by no means an expert on how people typically perform the sort of mixed digital/RF design you seem to be aiming at here, but my understanding is that many companies don't have the resources to be able to perform completely "integrated" simulations with any one tool. Instead various "point" tools are used to simulate the pieces, and if the interfaces between them have been defined well enough, there's a decent chance that everything will work once it's all plugged together. The various types of simulators used include:

-- Linear frequency domain simulators: These programs (stuff like Agilent's EESOF or Eagleware's Genesys linear simulator... oops, wait, Agilent now owns Eagleware too...) accept component data in the form of S parameters and let you quickly perform functions such as filter design, matching networks, power supply decoupling design, etc. They're also good for small signal amplifier design, although they require you to pick an operating point first and somehow come up with the S paramaeters of, e.g., a transistor that corresponds to that operating point.

-- Non-linear frequency domain simulators: These programs -- dominated by so-called "harmonic balance" simulators -- figure out the steady-state response of a non-linear network to a single tone (or small handful of tones) input. This makes it very easy to get the first draft of a mixer or non-class A amplifier going, and the results are computed orders of magnitude faster than what SPICE can do.

-- System simulators: These programs -- such as Systemvue by Elanix, now owned by Eagleware/Agilent -- model all your components "parametrically." For something like an amplifier, you just input items such as gain, noise figure, excess noise vs. frequency, etc. For mixers you provide the input-output responses (e.g., how much of the signal at the mixing port is passed through), etc. In general you get to provide a transfer function in the form of a polynomial, and the simulator takes care of the rest, computing items such as noise margins, conversion gains, etc. at some specified output port. This sort of "simulation" can actually be performed pretty well with an Excel spreadsheet, although a tool designed for the task tends to be a lot easier to set up and obtain results from.

-- Signal integtiry simulators: For items such as your FPGA, typically all you'll get out of the vendor is a so-called IBIS model. IBIS models characterize the input-output characteristics of the I/O cell of your FPGA in the form of tabulated or curve-fitted data around a handful of "bias" points; this is sufficient to allow you to check items such as eye patterns and setup and hold times of digital signals running around your board. Hyperlynx used to be one of the well-known, low-priced simulators here, although they're now owned by Mentor graphics, which usually indicates at least another zero was appened to the price tag.

-- Field solvers: The idea here is to take a relatively arbitrary physical structure in a defined environment (known dielectrics, known conduction of the materials of the metals involved, etc.) and extract a model of the device to use in (typically) a linear simulator. These are used all the time if you're designing spiral inductors on ICs, antennas, interconnects (bond wires, vias, BGAs, etc.), etc. HFSS is one of the heavyweights here, although there are plenty of more restrictive (2D and "2.5D") field solvers that often work fine (Sonnet, Agilent's Momentum, etc.) and have nowhere near the price tag or learning curve of something as sophisticated as HFSS.

Unless you're trying to squeeze the last once of performance for some design that you're going to manufacture and make a million of anyway, I'd suggest you perform a system level simulation (this will force you to spend time getting familiar with the parts' datasheets) and then just go and build your board and see what happens. It appears that you're performing board-level design using discrete RF "building blocks," and as such effects such as the PCB material you use, your layout technique, etc. will have a small but noticeable effect on the operation of the device and -- if this is the first time you're doing it -- it's unlikely you'll be able to properly account for all of these effects in your simulation.

If you do want to try out some of the fancier simulations, note that most companies such as Agilent, Ansoft, Sonnet, etc. will either give you or sell you a fraction of the usual cost their products if you're associated with an educational institution. Before persuing that route, you might want to sit down and play with some of the tools that are completely free, such as Sonnet Lite, RFSim99, etc.

---Joel Kolstad

These are the parts

Thank you for replying to our group's cry for help. Yes, we most recently became aware of the difficulties surrounding simulation of RF. We appreciate your comprehensive reply.

So we are trying to prototype our RF circuit. do you think its a good idea to try and use wire wrap or go strait to pcb? we havent got a clue what kind of trace widths to use. all i know is we have a circuit designed, and the parts are really really small.