Feedforward Control of Data Rate in Wireless Networks

现代无线网络使用发射功率的反馈控制来适应不断变化的信道条件，如传播损耗、阴影、多用户干扰等。为了更有效地对抗这些干扰，并适应信道模型的不确定性，本文建议在现有的功率控制方案之外，利用数据速率的前馈控制。这些控制器使用在前一个数据包中观察到的误码频率来计算下一个要传输的数据包的数据速率。我们的初步结果表明，理想情况下，这种方法至少可以提高20%的吞吐量，而不会增加额外的功率支出，或者至少降低30%的传输功率，而不会降低吞吐量。在某些场景中，这种方法可以将吞吐量提高300%或节省600%的电力。此外，这种方法的有效性与是否考虑单用户或多用户环境无关，并且不会发生数据速率战争。前馈数据速率控制的有效性主要有以下两个原因：(a)与反馈功率控制不同，反馈功率控制由于功率增加/减少的步长有限，适应速度相对较慢，而前馈数据速率控制是在一个数据包传输时间的范围内进行适应的。这导致更有效地拒绝快速干扰，如阴影水平。上述最小的吞吐量增加和功耗降低是由于这种快速适应能力。(b)与不适应信道不确定性（例如，信道是AWGN还是Rayleigh）的反馈功率控制不同，前馈数据速率控制能够适应这些影响。上述吞吐量提高三倍，功耗降低六倍，正是由于这一事实。本提案考虑的前馈数据速率控制器的开发方法基于以下三个步骤：(i)首先，设计一个非因果和非现实但最优的前馈数据速率控制器。它是非因果的，因为它将最佳数据速率计算为待传输数据包中误码概率的函数。这是不现实的，因为它使用的是误码的概率，而不是这个事件的频率。它的设计仅仅是为了得到可实现吞吐量的最小上界。（ii）接下来，这个控制器的原因和现实。通过使每个数据包的数据速率成为前一个数据包中的误码概率的函数来实现因果关系。通过使用误码的频率而不是误码的概率来实现。从而得到了一种可实现控制器，并对其性能进行了评价。结果表明，对于所有实际速度的手机，因果关系导致的性能下降相对较小。但是，使用频率而不是概率可能会导致严重的性能损失。因此，必须对误码频率进行一定程度的滤波。（iii）最后，介绍了上述可实现控制器的滤波版本，并证明了适当的滤波水平可以提高性能。在这一点上，这种级别的滤波仅在实验上（即数值上）进行了研究，并且设计正确滤波器的严格方法以及其他方法是在拟议的研究中要解决的问题。根据上述研究结果，本文拟开展的主要研究任务如下：1.研究方法：开发可实现的无线网络前馈数据速率控制器的设计方法。当使用该技术时，量化吞吐量增加和/或传输功率降低的水平。开发一种体系结构，在该体系结构中，前馈数据速率控制可以在蜂窝和ad-hoc环境中使用。所提出的研究的影响是为无线网络设计者提供了一种对抗信道干扰和不确定性的新方法。

Modern wireless networks use feedback control of transmit power to accommodate changing channel conditions, such as propagation loss, shadowing, multi-user interference, etc. This proposal suggests the utilization of feedforward control of data rate, in addition to the existing power control schemes, in order to more effectively combat these disturbances and, in addition, accommodate channel model uncertainties. These controllers use the bit error frequency, observed in the previous packet, in order to calculate the data rate of the packet to be transmitted next. Our preliminary results indicate that ideally this approach leads to a minimum of 20% throughput improvement, without additional power expenditures, or to a minimum of 30% decrease of transmit power, without decreasing the throughput. In some scenarios, this approach may lead to as much as 300% of throughput improvement or to 600% of power saving. Moreover, the efficacy of thisapproach is independent of whether a single or multi-user environment is considered, and no data rate wars take place. The efficacy of feedforward data rate control is mainly due to the following two reasons:(a) Unlike feedback power control, which adapts relatively slowly due to a finite step of power increase/decrease, feedforward data rate control adapts in the span of one packet transmission time. This leads to a more effective rejection of fast disturbances, such as the level of shadowing. The above-mentioned minima of throughput increase and power decrease are due to this fast adaptation capability.(b) Unlike feedback power control, which does not adapt to channel uncertainties (e.g., whether the channel is AWGN or Rayleigh), feedforward data rate control does accommodate these effects. The above-mentioned three-fold in-crease of the throughput and six-fold decrease of power are exactly due to this fact.The approach to the development of feedforward data rate controllers, considered in this proposal, is based on the following three steps: (i) First, a non-causal and non-realistic but optimal feedforward data rate controller is designed. It is non-causal because it calculates the optimal data rate as a function of bit error probability in the packet yet to be transmitted. It is non-realistic, because it uses the probability of bit error rather than the frequency of this event. It is designed solely in order to derive the least upper bound of the achievable throughput.(ii) Next, this controller is causified and made realistic. The causification is achieved by making the data rate of each packet a function of bit error probability in the previous packet. It is made realistic by using the frequency of bit error rather than its probability. Thus, an implementable controller is obtained and its performance is evaluated. It is shown that causification leads to a relatively small decrease of performance for all practical speeds of mobiles. However, using frequencies instead of probabilities may lead to a substantial performance loss. Thus, a certain level of filtering of bit error frequency is necessary.(iii) Finally, a filtered version of the above implementable controller is introduced and it is shown that a right level of filtering leads to an efficient performance. At this point, this level of filtering is investigated only experimen-tally (i.e., numerically), and a rigorous method for designing right filters is, along with others, a problem to be addressed in the proposed research. Based on the results to-date briefly mentioned above, the main tasks of the proposed research are as follows:1. Develop methods for design of implementable feedforward data rate controllers for wireless networks.2. Quantify the level of throughput increase and/or transmit power decrease when this technology is used.3. Develop an architecture in which feedforward data rate control can be used in both cellular and ad-hoc environ-ments. The impact of the proposed research is in providing wireless network designers with a new method for combating channel disturbances and uncertainties.