Automatic Modulation Classification in Mobile OFDM Systems with Adaptive Modulation

Adaptive modulation is an efficient way to combat the effects of deep fades in broadband orthogonal frequency division multiplexing (OFDM) systems with time-varying multipath channels, where modulation schemes are adapted to the current channel state. Bandwidth efficient modulation schemes are applied on subcarriers with high channel quality, while robust modulation schemes or even no modulation is preferred for subcarriers in deep fades. The resulting benefit in terms of required transmit power was demonstrated for a fixed data rate in the literature, where a gain of 5 ··· 15 dB was recorded for a BER of 10 -3 over the OFDM system with a fixed modulation. Alternatively, adaptive modulation can also be used to improve the bandwidth efficiency under the condition of a constraint transmit power and predefined quality of service (QoS). In the literature, several algorithms for adaptive modulation have been proposed with different emphasis on bandwidth efficiency and implemental complexity. In the thesis, the algorithm proposed by Chow is used due to its excellent tradeoff between performance and complexity.

A main drawback of adaptive modulation is that it requires a synchronization of adapted modulation schemes to enable demodulation in the receiver. Traditionally, this synchronization is organized in forms of explicit signalling, which reduces the bandwidth efficiency due to the signalling overhead. In the thesis, proposals are developed to reduce this undesirable overhead. These proposals exploit the correlation properties inherently existing in the transmission channel in both time and frequency domain, which leads to memory effects in the signalling source. State-dependent Huffman coding schemes are then applied to reduce the redundancy resulting from these memory effects.

This signalling overhead can be totally eliminated by automatic modulation classification (AMC). In the past, AMC was mainly of interest in military fields like threat analysis and electronic surveillance, where no prior knowledge about the used modulation scheme is available. The received signal is the single information source for classification. Under such circumstance, maximum likelihood (ML) based AMC provides the optimum solution in the sense that the classification error probability is minimized. Nowadays, AMC is drawing more and more research interest also in civilian applications like systems with adaptive modulation, where certain co-operations are organized as in the system considered in this thesis. These co-operations provide certain prior information, which can be utilized to improve the classification reliability. Consequently, the ML based approach does not deliver the minimum error probability any more. Investigations have to be conducted to verify how much the performance can be improved by incorporating this prior information into the AMC algorithm. As one focus in this thesis, a maximum a posteriori (MAP) based AMC is developed, which is potentially able to minimize the classification error probability again. Another focus is to reduce the implemental complexity to enable the application of AMC in systems with high time requirements like real-time systems.

In the last part of the thesis, comparisons are performed between these two approaches, namely explicit signalling and signalling-free AMC, in terms of final packet error probability. To ensure a fair comparison, the net data rate is always maintained as a constant in both operation modes.