The strapdown inertial navigation system (SINS) is employed for long-distance positioning of coal mine roadheaders due to its unique advantages. However, SINS mounted on roadheader is inevitably affected by external factors such as carrier motor vibrations and cutting-induced vibrations. To address this issue, we propose DiffSINS, a self-guidance Denoising Diffusion Probabilistic Model for one-dimensional SINS signal denoising. This non-autoregressive model transforms clean signals into random Gaussian signals through a Markov chain with constant synthesis steps, and is effectively trained by optimizing a variant of the variational lower bound of data likelihood. DiffSINS integrates Mamba and Transformer modules via a gated activation unit, coupled with residual connections and time embeddings, enabling efficient segmentation and extraction of noise features from one-dimensional time series data. Experimental validation is conducted using 6-dimensional raw noisy SINS data affected by vehicle body vibrations and simulated datasets. Results demonstrate that DiffSINS can accurately mitigate the impact of body vibrations on signals, preserve the fundamental frequency characteristics of SINS data, and achieve high quality reconstruction of vibration-affected data. Moreover, compared with the original noisy data, using DiffSINS to denoise the navigation data can improve the position estimation performance by 44%.
1 January 2026 in IEEE Transactions on Instrumentation and Measurement