Bluetooth Low Energy (BLE) is one of the most popular wireless protocols for building IoT applications because of its low energy, low cost, and wide compatibility nature. However, BLE communication performance can be easily affected by blockages because of its low transmission power. This paper presents BLEW, a technique to improve the BLE communication performance over weak links by exploiting adaptive symbol extension to combat channel interference and maximize network throughput. This paper addresses two key challenges. First, we propose a phase peak clustering-based approach for accurate preamble detection. It exploits the phase difference between sample points and coherently adds up the phase difference over the whole preamble to form a cluster with prominent phase peaks. After that, we propose a multi-domain DNN-based demodulator to enhance the symbol demodulation performance with low Signal-to-Noise Ratio (SNR). It fully extracts the temporal and spectrum features of the signal and effectively decodes BLE symbols. Furthermore, we model the BLE throughput considering the physical layer details of Commercial Off-The-Shelf (COTS) BLE chips, which can be used to choose the optimal symbol length according to the channel conditions. We implement BLEW with USRP B210 and COTS Nordic nRF52840 platform. The experiment shows BLEW can increase throughput up to 149.78 Kb/s compared with the native BLE. The phase peak clustering-based preamble detection method has an 8.27%-19.15% higher preamble detection rate and the DNN-based demodulator can bring up to 3.36 dB gain compared with existing methods.