<p>Intelligent reflecting surface (IRS) dynamically reconfigure the wireless propagation environment and have proven to be an effective approach for improving signal coverage, reliability, and network performance. Achieving optimal performance in IRS systems heavily depends on accurate channel state information (CSI), which is challenging because of high dimensionality of the cascaded channels and passive nature of IRS elements. Traditional channel estimation (CE) methods often struggle with these challenges, leading to increased computational complexity and reduced estimation accuracy. To mitigate these issues, a hybrid deep convolutional bidirectional long short term memory (CBLSTM) based CE is proposed for an IRS supported single input multiple output (SIMO) system employing orthogonal frequency division multiplexing (OFDM). The performance of the proposed model is compared with conventional MMSE method, long short term memory (LSTM), convolutional neural network (CNN), hybrid CNN-LSTM, and bidirectional LSTM (BiLSTM). The effectiveness of each model is evaluated using mean squared error (MSE), mean absolute error (MAE) metrics, signal to noise ration (SNR) gain and spectral efficiency. Simulation findings confirm that the proposed CBLSTM model achieves superior performance exhibiting the lowest estimation error, with an MSE of -7.00 dB, MAE of 0.43, and an SNR gain of 1.412 dB at 20 dB SNR, outperforming all other deep learning-based methods. The effectiveness of proposed hybrid deep learning framework results in delivering accurate and efficient CE for IRS supported OFDM systems.</p>