
From:  Prof. Dr.Ing. Heinrich Warmers 
Subject:  Re: [Paparazzidevel] IMU's 
Date:  Wed, 10 Aug 2011 12:20:59 +0200 
Useragent:  Mozilla/5.0 (Windows; U; Windows NT 5.0; deDE; rv:1.4) Gecko/20030619 Netscape/7.1 (ax) 
Hi Chirstophe, i agree with yours comments. We used the true airspeed sensor, an have be flown up to a wind of 15m/s. The delta flow up to 28m/s. Is this a low speed? We use the DCM rate of 25 Hz the same as the feedback loops. On HD videos taken from a camera on the aircraft we were surprised that the fly was verry smooth. We saw no ozillations. We saw ozillations, when we used a low roll angle 40° border. After setting this value to 80° the behavior was ok and the aircaft flow tunrs wiht a 20 m radius. Fore the Implementation of the DCM code there are 2 main points. 1. Correction to hold the orthogonality of the DCM matrix in every calculation step 2. Correction of the attitude also when flying circles The run time of each step is less than 1.5 ms for floating point calculation. Most of the time is spend in calculation of the transcendental angle functions. This can be speed up by using tables as Paparazzi does for fix point calculations. Since the filter has a complementary filter structure, you can all ways find situations in witch the filter fails. But this is also true for full kalman filters. My idea to have a correct attitude information is to use different ranges of the accelerator sensors. We have situations in which the 10 bit resulution of the adxl355 or adxl354 is bad. 10 bit of a +3g sensor give a pure angle error of +0.3°. This seems to be to high even for quadrocopters. When we fly wiht poylon electric aricraft we have acellerations in the range of 10g. The problem can be solved by usenig 2 sensors (One full range 2g, and one full range 16g). Another point to high speed calculation for quadrocopters. Most quadrocopters use the gyro signal direct (without building a difference ) to get stability, called back stepping. For this a high rate ( 50 Hz) is necessary. The main point for high frequencies is to deal with Shannon theorem even the bandwidth is to high. The motors generate vibrations in the range of 200..2000 Hz. These frequencies are mixed in the gyro signal. I measured this signals has a amplitude up to 0.3 of the maximum rate signal. Another point is the calculation of the non linearity. Same algorithms need very small delta angle to get the the fix point of the solution. My own quadocopter fly with a feedback frequency of 20 Hz. and a gyro signal rate of 100Hz.. Another club members use 50 Hz on a AVR (quads and three copters 20cm diameter), both without problems. So i do not understand why many projects (ASTEC, Paparazzi, UAVP, Openpilot etc.) deal with very high rates. Regards Heinrich Christophe De Wagter schrieb:

[Prev in Thread]  Current Thread  [Next in Thread] 