Consider multicell heterogeneous networks (HetNets), where each cell consists of a number of small (micro/pico) base stations (SBSs) and the SBSs perform coordinated multipoint (CoMP) transmission/reception to serve a number of downlink/uplink (DL/UL) users over the same frequency band. The SBSs work in full-duplex (FD) mode and the DL/UL users are half-duplex (HD). Due to co-channel transmissions, the DL and the UL are mutually interfered. This motivates us to consider a joint DL/UL beamformer design for maximizing the system sum rate. Different from the conventional sum rate maximization (SRM), where the user's rate is allowed to be any nonnegative real number, herein we focus on the discrete-rate case — each user's rate is constrained to a prescribed discrete-rate set. This discrete-rate constraint is motivated by the fact that practical communication systems are designed for a finite number of transmission rates, owing to the finite number of modulation and coding schemes (MCS). In view of this, our SRM problem is coined as discrete SRM (DSRM). The DSRM problem is essentially a mixed-integer nonlinear program, and NP-hard. To tackle it, two different approaches are proposed. The first approach is built upon the simplex reformulation of the discrete rate and the mirror descent (MD) algorithm. The second approach exploits the correspondence between the rate and minimum mean-square error (MMSE), and employs the block-coordinate descent (BCD) method to find an approximate solution. Simulation results demonstrate the superior rate performance of the proposed designs as compared with the weighted MMSE (WMMSE) design with naive rate quantization.

中文翻译：

---

全双工多小区异构网络的联合上行/下行离散和速率最大化

考虑多小区异构网络 (HetNet)，其中每个小区由多个小型（微/微微）基站 (SBS) 组成，SBS 执行多点协作 (CoMP) 传输/接收以服务于多个下行链路/上行链路 (DL/ UL) 同一频段的用户。SBS 工作在全双工（FD）模式，DL/UL 用户是半双工（HD）。由于同信道传输，DL和UL相互干扰。这促使我们考虑联合 DL/UL 波束形成器设计以最大化系统总速率。不同于传统的总速率最大化（SRM），其中允许用户的速率是任何非负实数，这里我们关注离散速率的情况——每个用户的速率被限制在规定的离散速率集合中。由于调制和编码方案 (MCS) 的数量有限，实际通信系统是为有限数量的传输速率设计的，这一事实激发了这种离散速率约束。鉴于此，我们的 SRM 问题被称为离散 SRM (DSRM)。DSRM 问题本质上是一个混合整数非线性程序，并且是 NP-hard。为了解决这个问题，提出了两种不同的方法。第一种方法建立在离散速率和镜像下降 (MD) 算法的单纯形重构之上。第二种方法利用速率和最小均方误差 (MMSE) 之间的对应关系，并采用块坐标下降 (BCD) 方法来寻找近似解。