Certain embodiments of the present disclosure provide techniques for approximate computation of l2 norms as a part of the maximum likelihood (ML) detection: tri-maxmin, maxsum and sortsum algorithms. The proposed approximation schemes show better accuracy than conventional approximation schemes—the abssum and maxmin algorithms, while the computational complexity is very similar. The error rate performance of the ML detection that utilizes proposed norm-approximation techniques are very close to the reference ML detection with exact calculation of l2 norms, while the computational complexity is significantly smaller.