The Research And Design Of An Easy Implemented Integrated Temperature Compensation Technique For MEMS Accelerometer

With the advancements of robotics,aviation,etc.the requirements of small-size,sensors products are in vast demand.MEMS accelerometer is one of the important sensors used in navigation.However,due to the expension and contraction nature of its structure,it is very vulnerable to temperature change in the working environment.The output signal of MEMS accelerometer will drift with the change of temperature.This temperature drift results in MEMS accelerometer output signal error and reduces the signal accuracy.Conventionally,the analog compensation method mainly uses components that have similar or opposite temperature properties to achieve an output that is invariant to temperature change.The disadvantage is that the temperature coefficient of components can be different in nature.So,it needs vast iterative simulations to achieve the optimal compensation.The digital compensation method usually uses model-based algorithm such as polynomial,neural network and Kalman filter etc.to compensate the output errors due to temperature change.The main disadvantage of the polynomial method is that it requires high orders compensation that leads to complex operations and poor stability,while the shortcomings of using neural network and Kalman filter algorithm are that they often need external micro-controller for operation control and are too complex and costly to be integrated into silicon.To solve the above shortcomings,this thesis presents an easy integrated temperature compensation technique that comprises a new function compensation model,a fitting algorithm and a hardware implementation method.The new function compensation model uses double exponential function to solve the complexity of implementing high order compensation.However,there exists a problem of low fitting precision and non-convergent issues using double exponential function.This thesis proposes a method to obtain a good seed for the fitting operation to avoid such issues.CORDIC algorithm is adopted to implement the double exponential function,however,there is range limits requirement for the data to be processed.This thesis proposes an approach to normalize the data within the range limits in order to avoid undefined values that is out the range.In addition,this approach is targetted towards ease of ASIC integration.A new method of CORDIC algorithm implementation based on state machine is designed,which can meet the requirement of speed and reduce the use of resources.The propose technique is designed and simulated in MATLAB.The result shows that the fitting effect is better than 6^th-order polynomial compensation method.The simulation also shows that the relative error of the calculated results between the digital circuit realization and the theoretical value is less than 10^-4.Temperature cycle experiment is performed on two accelerometers and measurement data are obtained.The proposed method is implemented on FPGA,and uncompensated and compensated measurement results are compared.The simulation results show that the maximum temperature drift reduces from 0.1023V before compensation to 0.13243mV after compensation,and from 0.0629V to 0.41319mV respectively;the peak values of the output temperature curve are reduced from 1.8513mV to 0.26223mV and from3.9439mV to 0.82108mV respectively.The results also show that the readings from FPGA are similar to that of the values calculated by MATLAB.