Stanford Research Communication Program
  Home   Researchers Professionals  About
Archive by Major Area


Social Science

Natural Science

Archive by Year

Fall 1999 - Spring 2000

Fall 2000 - Summer 2001

Fall 2001 - Spring 2002

Fall 2002 - Summer 2003




Power Control in Wireless Communications

Tim Holliday
Management Science and Engineering
Stanford University
June 2001

One of the key trade-offs in mobile wireless communications is power consumption vs. performance. In many cases, phones used for high bandwidth activities run out of battery power very quickly. Furthermore, a mobile phone transmitting at high power can interfere with other transmitters. My research is focused on power control algorithms that simultaneously provide a guaranteed level of quality of service and greatly reduce power consumption, resulting in a much better experience for the user and a substantial increase in battery life.

The problem of power control for wireless communications has been well studied. Consider the typical setup of a group of mobile devices transmitting data to a base station. These mobile devices are faced with time-varying wireless channels, where the path loss in the channel and interference from other users changes randomly over time. As the path loss or interference increases the probability of a mobile device successfully transmitting data goes down. Or put another way, think of trying to hold a conversation with a friend in a crowded room your voice is the mobile transmitter and your friend's ear is the base station. Interference is like the voices of other people in the room; if they are speaking at a high volume your friend will not be able to distinguish your voice. Path loss, on the other hand, results from the appearance of objects (e.g. a vase, table, or door) between you and your friend. Of course, in the context of wireless communications, path loss is caused by much larger objects like hills, buildings, and so forth. If the channel conditions (path loss and interference) in the crowded room are poor, you can attempt to communicate with your friend by shouting, or by using very simple words or hand signals. Another option is to wait for everyone else to quiet down or move to another part of the room. This is analogous to what we try to do for wireless devices if conditions are poor, we can raise the transmitter power (start shouting), reduce coding complexity (use simpler words), or withhold transmission until the channel improves.

Solutions to this problem have been proposed using many different mathematical frameworks, but they tend to use the same approach reduction of power and coding complexity by the mobile as it waits for a clear channel. Of course, while the mobile is waiting for the channel to clear, it also has data waiting for transmission. This policy of halting transmission during poor channel conditions adds substantial amounts of time delay to data, resulting in unacceptable performance from the mobile user's perspective. Clearly, there is a trade-off between providing some guaranteed level of performance while minimizing power consumption. In some cases, the mobile should lower its transmit power during poor channel conditions to conserve power, and in other cases it should raise transmit power in order to satisfy a certain quality of service constraints. How do we solve this problem?

My research is focused on control policies for data that is sensitive to delay (e.g. voice, video, and multimedia). The control problem is to minimize the average transmitter power subject to a constraint on the probability distribution of lost data. This constraint on the distribution of data loss, rather than simply the total data lost, is very important. In the past, many control policies have only constrained average delay or the total probability of data loss. For example, suppose you only constrain the probability of data loss to be less than five percent. Over 10,000 data packets, this constraint is satisfied if you drop five out of every 100 packets. However, it is also satisfied if you transmit 9,500 packets successfully and then drop the last 500. Clearly, the latter case is unacceptable, which is why a constraint on the total distribution of data loss is desirable.

A dynamic program can solve the dilemma of ensuring acceptable data loss while minimizing power consumption. The idea behind the dynamic programming theory is to determine what actions the mobile should take when faced with a problem that has a random outcome. In this case, the successful transmission of data, the changes in the wireless channel, and the changes in data arriving at the transmitter are all random processes. By solving an appropriately formulated dynamic program, we can discover control policies that greatly reduce transmitter power and satisfy the constraint on the distribution of data loss.

The specific formulations for some examples are beyond the scope of this paper, but they are available in our publications. When compared with the industry standard policies and other policies proposed in academia, we show a significant improvement in both quality of service and power consumption. In many cases, our methodologies provide a reduction of 50 percent or more in the average power used for transmission. Depending on the battery and radio technology used, battery drain is exponentially proportional to the transmitted power. Therefore, a reduction in transmitted power of 50 percent can more than easily double the life of a battery.