Kalman Filters

I mostly use C# for desktop style applications, but lately I'm in need of an embedded controller. The inputs I'm working with tend to contain a fair amount of noise, so I'll need to develop some sort of smoothing algorithm. Normally I would just use a Kalman filter, but I'm doubting the performance I'll be able to harness from an embedded environment for realtime operation. Has anybody implemented a Kalman filter on the Netduino that runs with good performance? Is there anything you had to optimize to squeeze performance out of the filter (i.e. fixed point arithmetic)? What have you found works the best?

I was just looking at porting some Kalman filters from C over to C# to run on the Netduino - however some of the System.Math functions are not available in .NETMF so it's going to take a bit of work. It might be worth implementing them in native code?

Has anybody implemented a Kalman filter on the Netduino that runs with good performance?

Planning to ... but just started !

Mike.

If you mean a sort of signal processing, then forget the C#. I have tried but it is absolutely unpredictable as timing, and terribly slow. Anyway, it seems that working with floats instead of ints, is performing almost the same. That is the only good info. I am trying to move some step in C/C++ to implement some library for real-time processing. I guess it will take a long time. Cheers

I was just looking at porting some Kalman filters from C over to C# to run on the Netduino - however some of the System.Math functions are not available in .NETMF so it's going to take a bit of work. It might be worth implementing them in native code?

There's a good supplementary maths library here.

There's a good supplementary maths library here.

The effort is really appreciable, but it is not enough.
Doing math in a managed way it results a very poor performance. I am fighting against a FFT and it takes almost 100ms to get the result (*VERY* simplified and with tons of hardness).

Another point over that particular library is to the square-root function I need to compute the modulus of a complex number. That algorithm is *really* slow.
Take a peek my article, if you interested in...
http://highfieldtale...he-square-root/

I'd expect to make simple calculations within microsecs, not millisecs.

Cheers

I am normaly working on automation project where I sometimes are using an algorithm from "Principles and Practices of Automatic Process Control" by Smith & Corripio. Perhaps some of you can use this quite simple aproach. I have only tried i out here in my workshop and i don't know if it will work in real life. But for me it seems to work fine. here is the Class: using System; using Microsoft.SPOT; /******************************************************************************** * Digital filter for noisy analogue signals. This equation is a standard * exponential filter taken from "Principles and Practices of Automatic Process * Control" by Smith & Corripio * * Equation: FltrPv = alpha * OldFltrPv + (1 - alpha) * RawPv * * where: * alpha = (tau / (tau + st)) * tau = time in sec for output to reach 63% of RawPv (Damping_time) * st = scantime, also update rate of equation * RawPv = variable to be filtered * FltrPv = filter variable * OldFltrPv = FltrPv (st - 1) ********************************************************************************/ namespace Netduino_PT100 { class Filter { const Int64 ticks_per_msec = System.TimeSpan.TicksPerMillisecond; float tau; float oldFltrPv; DateTime oldScantime = DateTime.Now; public Filter(float Tau = 10.0F) { tau = Tau; } public float Tick(float RawPv) { float st = (DateTime.Now.Subtract(oldScantime).Ticks / ticks_per_msec) * 0.001F; oldScantime = DateTime.Now; float alpha = (tau / (tau + st)); oldFltrPv = alpha * oldFltrPv + (1 - alpha) * RawPv; return oldFltrPv; } public void Set_tau(float Tau) { tau = Tau; } } } and here is the main program: using System; using System.Threading; using Microsoft.SPOT; using Microsoft.SPOT.Hardware; using SecretLabs.NETMF.Hardware; using SecretLabs.NETMF.Hardware.Netduino; namespace Netduino_PT100 { public class Program { public static void Main() { AnalogInput ain_1 = new AnalogInput(Pins.GPIO_PIN_A0); ain_1.SetRange(-40, 1040); Filter ain_1_filter = new Filter(); while (true) { Thread.Sleep(200); Debug.Print("Temperatur Filter:" + ain_1_filter.Tick(ain_1.Read() * 0.1F).ToString("N2")); Debug.Print("Temperatur Raw :" + (ain_1.Read() * 0.1F).ToString("N2")); } } } } /dennis

better not use NETMF for signal processing or anything that involves heavy math iteration. It's just too slow.