Signal Generator

Use pretty much any processor you like - the cost will be dominated by the FPGA, clock generator/DDS, DAC, reconstruction filter etc.

You can go out and buy signal generators like this and there is a reason they cost £2000 or more.

To get 100MHz sine wave at reasonable noise and distortion you will need at least 3e8 samples per second, even fast processors can't do IO at that rate so you'll need the FPGA to work the DAC.

Use a processor with the necessary features from the family you are most familiar with.

Come on Michael, thats a bit of a cop out , I cant use pretty much any processor I like because of the 100Mhz problem, I reckon if you get a fast enough processor the actual system is not that difficult to program, high speed DACs might be however....

John - you haven't quite got my point - I don't know of ANY processor that can drive a DAC at 300MHz with reasonable jitter. A DAC that fast will probably have LVDS inputs  - an easy interface to an FPGA but not to any processors I know.  Since you will need an FPGA (or other custom logic) the processor doesn't have any demanding real time work to do - so it is not a performance driven decision (at least as far as the signal generation goes). This is the case in a great many real development projects.

You haven't given anything like enough spec details to really drive the processor choice but if I were doing this (on the basis of the limited information) I would use an ARM Cortex, probably one of the STM32F parts, because I am familiar with them so the cost of development would be lower. If you are tooled up for a PIC32 or something else - that would probably do just as well.

Hi John,

I've used this type of solution before on a PIC processor so I have a little experience on the subject. I think it's going to be pretty difficult to achieve what you're looking for using fully digital means. Here's the issue, and again this is the way I've done it, perhaps someone can mention a better way.

Basically a produce a PWM waveform at the highest frequency my hardware will allow, or that I deem suitable. The idea is to use the PWM as a DAC. Fourier analysis tells us that a positive square wave (0 - 5V) will have frequency components at odd multiples of the fundamental as well a strong fundamental, additionally since it's a positive square wave we have a DC average value.

It's the DC value that allows us to create arbitrary waveforms. If we vary the pulse width we can vary the DC component, in order to obtain the desired waveform we need to filter out the harmonics, so in order to obtain a clean waveform you'll need probably a filter of 3rd or fourth order.

So basically you program the waveform by modulating the pulse width with the waveform. This method works very well for arbitrary waveforms. In my earlier experiment I got pretty amazing results just using a first order filter( Not the cleanest signals but much better than I expected).

Besides the fact that you have to get a high speed PWM going which you can modify cycle by cycle(Takes one heck of a processor). In order to represent arbitrary waveforms with a high degree of accuracy you would want to be able to have at least 100 points for every period of the waveform. To reproduce one point you need one PWM cycle so If you run your PWM at 100Mhz the highest frequency you could faithfully recreate with this method would be a 1Mhz arbitrary signal.

Following this same reasoning if you wanted to produce an arbitrary signal from 0-100MHz you would need a 10GHz PWM waveform! I'm not aware of too many computers that can run that fast let alone an embedded processor.

Now that I've shattered your will to live, here's the good news. You could cut the 100 points per period minimum to 50 points. You mentioned you only want to create Sine, Triangle, and adjustable Square wave so the ability to create an arbitrary waveform is not really necessary.

Another more feasible solution would be to use PLL which could easily generate a 100MHz Square wave, if you filter it you'll get a 100 MHz sine wave and with some other processing you can get a triangle waveform. I'm not to clear on the details but I believe many commercial units use this method.

Hi John,

I've used this type of solution before on a PIC processor so I have a little experience on the subject. I think it's going to be pretty difficult to achieve what you're looking for using fully digital means. Here's the issue, and again this is the way I've done it, perhaps someone can mention a better way.

Basically a produce a PWM waveform at the highest frequency my hardware will allow, or that I deem suitable. The idea is to use the PWM as a DAC. Fourier analysis tells us that a positive square wave (0 - 5V) will have frequency components at odd multiples of the fundamental as well a strong fundamental, additionally since it's a positive square wave we have a DC average value.

It's the DC value that allows us to create arbitrary waveforms. If we vary the pulse width we can vary the DC component, in order to obtain the desired waveform we need to filter out the harmonics, so in order to obtain a clean waveform you'll need probably a filter of 3rd or fourth order.

So basically you program the waveform by modulating the pulse width with the waveform. This method works very well for arbitrary waveforms. In my earlier experiment I got pretty amazing results just using a first order filter( Not the cleanest signals but much better than I expected).

Besides the fact that you have to get a high speed PWM going which you can modify cycle by cycle(Takes one heck of a processor). In order to represent arbitrary waveforms with a high degree of accuracy you would want to be able to have at least 100 points for every period of the waveform. To reproduce one point you need one PWM cycle so If you run your PWM at 100Mhz the highest frequency you could faithfully recreate with this method would be a 1Mhz arbitrary signal.

Following this same reasoning if you wanted to produce an arbitrary signal from 0-100MHz you would need a 10GHz PWM waveform! I'm not aware of too many computers that can run that fast let alone an embedded processor.

Now that I've shattered your will to live, here's the good news. You could cut the 100 points per period minimum to 50 points. You mentioned you only want to create Sine, Triangle, and adjustable Square wave so the ability to create an arbitrary waveform is not really necessary.

Another more feasible solution would be to use PLL which could easily generate a 100MHz Square wave, if you filter it you'll get a 100 MHz sine wave and with some other processing you can get a triangle waveform. I'm not to clear on the details but I believe many commercial units use this method.