The HQC encryption framework is a general code-based encryption scheme for which decryption returns a noisy version of the plaintext. Any instantiation of the scheme will therefore use an error-correcting procedure relying on a fixed auxiliary code. Unlike the McEliece encryption framework whose security is directly related to how well one can hide the structure of an error-correcting code, the security reduction of the HQC encryption framework is independent of the nature of the auxiliary decoding procedure which is publicly available. What is expected from it is that the decoding algorithm is both efficient and has a decoding failure rate which can be easily modelized and analyzed. The original error-correction procedure proposed for the HQC framework was to use tensor products of BCH codes and repetition codes. In this paper we consider another code family for removing the error vector deriving from the general framework: the concatenation of Reed-Muller and Reed-Solomon codes. We denote this instantiation of the HQC framework by HQC-RMRS. These codes yield better decoding results than the BCH and repetition codes: overall we gain roughly 17\% in the size of the key and the ciphertext, while keeping a simple modelization of the decoding error rate. The paper also presents a simplified and more precise analysis of the distribution of the error vector output by the HQC protocol.