In the time past, virtually all the contemporary communication systems depend on distinct source and channel encoding schemes for data transmission. Irrespective of the recorded success of the distinct schemes, the new developed scheme known as joint source channel coding technique has proven to have technically outperformed the conventional schemes. The aim of the study is centered in developing an enhanced joint source-channel coding scheme that could mitigate some of the limitations observed in the contemporary joint source channel coding schemes. The study tends to leverage on recent developments in machine learning known as deep learning techniques for robust and enhanced scheme, devoid of explicit code dependence for the signal compression and as well in error correction but learn automatically on end-to-end mapping structure for the source signals.It primarily aimed at providing an improved channel performance approach for wireless communication network. A deep learning algorithm was implemented in the study, the scheme focused on improving the Bit Error Rate (BER) performance while reducing latency and the processing complexity in Joint Source Channel Coding systems. The deep learning autoencoder model was deployed to compare with the hamming code, convolution code, and uncoded systems. JSCC using neural networks were simulated based on BER performance over a range of energy per symbol to noise ratio (Eb/No). Training and test error for the fully connected neural network autoencoder models on channels with 0.0dB and 8.0dB were carried out. The results obtained showed that the autoencoder model had a better BER performance when compared with the convolution code and uncoded systems, it also outperformed the uncoded BFSK with an approximately equal BER performance when compared with the hamming code (soft decision) decoding system.