This article experimentally demonstrates a frequency-sweeping notch-beam sub-THz radar frontend based on a two-line array antenna featuring computational imaging. Operating within 237.5 GHz and 250 GHz with 12.5 GHz bandwidth, the radar utilizes a 12 λc delay line to achieve frequency-sweeping capabilities. This configuration allows dynamic notch-beam scanning across angular ranges from − 26.5 ∘ to 28 ∘ . The radar frontend is highly compact with a total size of 20 mm× 14.3 mm× 1.2 mm, including the beam-steering network, a magic-tee for creating the 180 ∘ phase shift required for creating the notch-beam, and the antenna array, and is implemented by silicon micromachining. The radar was evaluated with single and dual-target scenarios utilizing and benchmarking different computational imaging algorithms, i.e., matched filter (MF), fast iterative shrinkage-thresholding algorithm (FISTA), and multiple signal classification (MUSIC). It was found that the MUSIC algorithm outperforms MF and FISTA in range and angular resolution in single-target scenes, achieving a range resolution of 7.8 mm and an angular resolution of 15.7 ∘ , with detection errors of less than 6.6 mm and 3.5 ∘ , respectively. Although the MUSIC algorithm maintains reliable range resolution in dual-target scenarios, it performs poorly in providing angular information.