Sequential Recommender Systems (SRS) focus on modeling the sequential behaviors of users. By doing so, they are able to capture users’ interests that often change and develop over time. In recent times, Transformer-based SRS have demonstrated remarkable achievements. Nevertheless, research findings indicate that the self-attention mechanism within Transformer-based models is, in essence, a low-pass filter. As a result, it overlooks high frequency information, which may contain valuable and meaningful patterns of user interests. This realization spurs us to explore superior filtering technologies for SRS. Eventually, we identify Discrete Wavelet Transform (DWT), a well-known time-frequency analysis method from the digital signal processing domain. DWT proves to be effective in handling both low-frequency and high-frequency information. We have devised an adaptive time-frequency filter that utilizes the DWT technique. This filter breaks down user interests into multiple signals with varying frequencies and time characteristics, and is capable of automatically learning the weights of these signals. Moreover, we have developed DWTRec, a model specifically designed for sequential recommendation. It is entirely based on the proposed adaptive time-frequency filter. Thanks to the efficiency of the fast DWT technique, DWTRec theoretically exhibits lower time complexity and space complexity, and excels at modeling long sequences. Our experimental results show that our model outperforms the current state-of-the-art baseline models across datasets with diverse domains, different levels of sparsity, and varying average sequence lengths. In particular, when compared to previous models, our model shows a significant performance improvement as the sequence length increases, further highlighting an additional advantage of our proposed model.