The work presented in this thesis aims to improve the performance of the Fundamental PWM sensorless control technique by proposing a new way to estimate current derivatives in the presence of high frequency oscillations. The Fundamental PWM technique offers performance across the entire speed range (including zero speed). The method requires current derivative measurements when certain PWM (Pulse Width Modulation) active and null vectors are applied to the machine. However the switching action of the active devices in the inverter and the associated large dv/dt result in current and current derivative waveforms being affected by high frequency oscillations which prevent accurate measurement of the current derivative. Other approaches have allowed these oscillations to decay before attempting to take a derivative measurement. This requires that the PWM vectors are applied to the machine for a time sufficient to allow the oscillations to decay and a derivative measurement to be made (the minimum pulse width). On some occasions this time is longer than the time a vector would have normally been applied for (for example when operating at low speed) and the vectors must be extended and later compensated. Vector extension introduces undesirable current distortion, audible noise, torque ripple and vibration. In this thesis the high frequency oscillations and their sources are investigated and a method of using Artificial Neural Networks to estimate current derivatives using only a short window of the transient current response is proposed. The method is able to estimate the derivative directly from phase current measurements affected by high frequency oscillations and thus allows a reduction in the minimum pulse width to be achieved (since it is no longer necessary to wait for the oscillations to fully decay) without the need for dedicated current derivative sensors. The performance of the technique is validated with experimental results.