# Achieving Non-Zero Information Velocity in Wireless Networks

###### Abstract

In wireless networks, where each node transmits independently of other nodes in the network (the ALOHA protocol), the expected delay experienced by a packet until it is successfully received at any other node is known to be infinite for signal-to-interference-plus-noise-ratio (SINR) model with node locations distributed according to a Poisson point process. Consequently, the information velocity, defined as the limit of the ratio of the distance to the destination and the time taken for a packet to successfully reach the destination over multiple hops, is zero, as the distance tends to infinity. A nearest neighbor distance based power control policy is proposed to show that the expected delay required for a packet to be successfully received at the nearest neighbor can be made finite. Moreover, the information velocity is also shown to be non-zero with the proposed power control policy. The condition under which these results hold does not depend on the intensity of the underlying Poisson point process.

Non-Zero Information Velocity in PPP {aug} , \thankstextt2Research Supported in part by UGC center for advanced studies. \thankstextt3Research Supported in part by INSA young scientist award grant. class=MSC] \kwd[Primary ]60D05 \kwd[; secondary ]60F15, 60K30

wireless networks, SINR graphs, expected delay, information velocity

## 1 Introduction

Typically, nodes in a wireless network are separated by large distances and packets are routed from source to their destination via many other nodes or over multiple hops. Therefore, to understand the connectivity or information flow in a wireless network, a space-time SINR graph is considered. Such a graph models the evolution of the spatial as well as the temporal connections in the network. The space-time SINR graph is a directed and weighted multigraph that represents the most complete random graph model for wireless networks [1]. The SINR (signal-to-interference-plus-noise-ratio) is a ratio of the relative strength of the intended signal and the undesirable interference from simultaneously active unintended nodes of the wireless network. The SINR between any two nodes evolves with time and depends not only on the distance between the two nodes but also on the location of the other nodes in the network. At any time, a directional connection is established from a node at to another node at if the SINR from to is larger than a threshold. Such a connection represents the ability of node to deliver meaningful information to .

Let be a point process that specifies the location of the nodes of the network. For any , let be the set of nodes that are transmitting at time and be the set of nodes in receiving mode at time . Formally, the SINR from a node to a node is given by

(1) |

where is the distance based signal attenuation or path-loss function, is the transmitted power from at time , is the interference suppression constant, , are the space-time fading coefficients that model the loss (or gain) from node to at time due to signal propagation via a wireless medium, and is the variance of the so-called additive white Gaussian noise. By an abuse of notation, we will often use for , since the path loss is a function of the distance between and . The term in the denominator of (1) is referred to as the interference. Note that we do not include the nodes at in the interference term since transmission from node is the signal of interest and node is in receiving mode. is set to be zero at time if either node or if node . Define the indicator random variables

(2) |

where is arbitrary. The space-time SINR graph is defined to be the graph , where a directed edge exists from to if . Given , the location of the nodes is static, and the time evolution of the graph is entirely due to changes in the fading variables and the set .

In this paper, we consider a space-time SINR graph in which the location of the nodes is modeled as a homogeneous Poisson point process (PPP). Modeling location of nodes in a wireless networks as a PPP is quite attractive from an analytical point of view and has paid rich dividends in terms of finding exact expressions for several performance indicators such as maximum rate of transmission (capacity), connection probability, etc. [2, 3, 4, 5], that are hard to derive otherwise. A PPP node location model is well suited for modeling both the ad hoc networks, where large number of nodes are located in a large area without any coordination, as well as the modern paradigm of cellular networks [5], where multiple different layers of base-stations (BSs) (macro, femto, pico) are overlaid on top of each other, and the union of all BSs appears to be uniformly distributed.

Given , the stochastic nature of the fading coefficients and the set implies that the event is a random variable, and hence potentially, multiple transmissions are required for successfully transmitting a packet from node to . Repeated transmissions entail delay in packet transmission, and it is of interest to make the expected delay as small as possible. Another related quantity of interest is the information velocity, that is defined as the limit of the ratio of the distance between the source and the destination of any packet, to the total delay experienced by the packet to reach its destination successfully over multi-hops, as distance goes to infinity.

Expected delay and the information velocity are closely connected to the various notions of capacities in wireless networks, e.g., throughput capacity [2], transport capacity[2], delay-normalized transmission capacity [6, 7] etc., since all of them are measures based on the successful rate of departure of packets towards their destination. Finding the speed of information propagation is also related to first passage percolation [8, 9], and dynamic epidemic processes [10, 11, 12], however, the analysis in the space-time SINR graph gets complicated due to the presence of interference.

In the seminal paper [2], it was shown that with PPP distributed node locations (albeit for a somewhat simpler model), the per-node throughput (rate of transmission between any two randomly selected nodes) or information velocity tends to zero as the size of the network grows. The most general analysis on the expected delay and the information velocity has been carried out in [13] for a PPP-driven space-time SINR network. It is shown that with an ALOHA protocol, where nodes transmit independently of all other nodes with fixed power, the expected delay required for a packet to leave a given node and be successfully received at any other node in the network is infinite. Remarkably, this result is shown to hold even in the absence of interference and requires only the additive noise to be present. Moreover, the information velocity is also shown to be zero. These result have tremendous ’negative’ impact on network design, since it shows that essentially any packet cannot exit its source with finite expected delay.

Both the results from [13], viz., the infinite expected delay and zero information velocity, are attributed to the fact that with PPP distributed node locations, a typical node can have large voids around itself, that is, regions that contain no other nodes with high probability. In such a circumstance, a large number of retransmission attempts will be required to overcome the effect of additive noise and support a minimum SNR at any of the other receiving nodes. Consequently, the mean exit delay is infinite (when averaged over the realizations of the PPP) and the information velocity tends to zero.

One solution prescribed in [13] to make the information velocity non-zero is to add another regular square grid of nodes with a fixed density, in which case the nearest neighbor distance is bounded, and the information velocity can be shown to be non-zero. From a practical point of view it is rather limiting to assume the presence of such a regular grid.

Some work has been reported on finite expected delay together with a finite bound on the information velocity [14, 15, 1], under restrictive assumptions such as assuming temporal independence with the SINR model, i.e. interference is independent between any two nodes over time, no power constraint and temporal interference independence, and no additive white Gaussian noise, respectively.

In this paper, we propose a power control mechanism to show that the information velocity can be made non-zero for the space-time SINR graph with PPP node locations without any additional restrictive assumptions on the network. In [13], the information velocity is defined as the limit of the ratio of the distance between two points and to the time it takes for the packet to go from to as the distance tends to infinity. The packet simultaneously traverses multiple paths and the time taken is the first time the packet is received at . This makes the set-up somewhat complicated to work with and so the results in [13] are proved for delays averaged over the fading variables. In order to overcome this problem, we modify the definition of information velocity by specifying a random path along which a tagged packet will traverse the network. Allowing for a larger set of paths and picking the one that is optimal as done in [13] will only increase the velocity. Thus our result provides a lower bound on the information velocity. We describe our idea in brief here and the precise definition will appear later.

We track a tagged packet as it traverses the network via a conic forwarding strategy along a prescribed path that depends only on the realization of the underlying distribution of the nodes. Briefly stated, conic forwarding works as follows. At each node, the plane is partitioned into multiple cones, and each node transmits the packet at the head of its queue to its nearest neighbor in the cone that contains the packet’s destination until it is successfully received. We refer to such a cone as the destination cone. This conic forwarding idea circumvents the problem of forming nearest neighbor loops, since the packet always progresses towards its destination. This also allows us to exploit the various independences that exist across space and time. The speed of this tagged particle along the prescribed path is what we will refer to as the information velocity (the direction of motion being contained in the choice of the transmission cones). If is the distance that the tagged packet travels in time , then is the information velocity. The aim of this paper is to devise a power-control strategy under which .

The power control strategy works by nullifying the path-loss from a node towards its nearest neighbor in the destination cone. In particular, since the path-loss between node and its nearest neighbor in the destination cone is , the transmitted power is taken to be , where is a constant. Since the nearest neighbor in a PPP can be at arbitrarily large distances, we need . To compensate for the non-homogeneity in power used at each node, we modulate the transmission probability , such that equals the average power constraint at each node.

We wish to note that per se, power control is not a new concept in wireless communication. For instance, in CDMA type wireless network [16], power control is mandatory to overcome the near-far effect, where mobiles that are nearer to the base station have to continuously update their transmitted power so as not to severely limit the transmission from mobiles that are farther away. However, the specific strategy that we use in this paper has not been considered in the literature. Further the use and advantages of power control in large wireless networks with randomly located nodes has not been fully explored.

In terms of implementation, the conic forwarding needs no special effort, since the transmitter only adjusts the power according to the nearest neighbor distance in the destination cone, and the transmission is isotropic, i.e., does not require any information about the direction, circumventing the need for any special hardware, e.g., directional antennas etc. For power control, the transmitter needs to learn the distance to its nearest neighbor in the destination cone. Nearest neighbor routing [17, 18] is standard in wireless communication networks, where packets are forwarded to the nearest neighbors, which requires discovery of nearest neighbors as well as their distances, and hence our power control policy does not entail any new overhead.

Using conic forwarding strategy together with power control, we show that the expected delay to the nearest neighbor in the destination cone is finite provided . In addition, as the tagged packet traverses the network from one node to another along the path specified by the conic forwarding strategy, the sequence of time delays turns out to be a non-stationary sequence. In order to overcome this problem, we add additional (virtual) nodes to the network as the packet moves from one node to another. The nodes are added in such a way that the path along which the packet traverses in not affected. The interference experienced by the particle increases (and consequently reduces its speed) but the technique delivers for us a stationary sequence of delay times. This enables us to apply the ergodic theorem and infer with probability one that the information velocity is strictly positive for the stationary sequence, and hence for the original sequence. It is important to note that these virtual nodes are not really required in practice to achieve non-zero information velocity, but are only used as a analytical tool to upper bound the per-hop delays (via increasing the interference) and making them stationary across different hops.

## 2 System Model

Let be a homogenous PPP with intensity in modeling the location of the nodes of the network. The time parameter is assumed to be discrete (slotted). Let be a collection of independent exponentially distributed random variables with parameter . is the fading power from node to node in the time slot . The path loss between denoted by is given by

where and is arbitrary.

We assume that each node can only operate in a half-duplex mode, that is, in the time slot , a node is on (transmitter) or off (receiver) following a Bernoulli random variable , with . Let . The set of on (off) nodes in the time slot is denoted by ().

Let be cones with angle in with vertex at the origin, satisfying and for , as shown in Fig. 1. Without loss of generality, suppose that is symmetric about the x-axis and opens to the right. Let be translation of cone by . In the time slot , for a node , let be the cone that contains the final destination of the packet that it wishes to transmit. We call this cone as the destination cone. Denote the nearest neighbor of in the destination cone by . If the node at is on in time slot then it transmits with power . The key idea in this paper is to employ a decentralized power control scheme, that is, the functions depend locally on . The particular forms that these functions take are given by

(3) |

where , is a constant, and is the average power constraint. Note that , since . Thus, in each time slot, each node makes transmission attempts with transmission power proportional to the distance to its nearest neighbor in the destination cone to compensate for the path-loss to the nearest neighbor. The transmission probability is chosen so as to satisfy an average power constraint.

In Fig. 2, we illustrate the transmission strategy, where each node transmits to its nearest neighbor in the destination cone (shaded cone) with line thickness proportional to the transmit power, farther the nearest neighbor, larger the power. In prior work [13], with the ALOHA protocol, the functions and were assumed to be constants that were independent of the system parameters.

Thus, the SINR from node to node in time slot is given by

(4) |

where is the processing gain of the system (interference suppression parameter) which depends on the transmission/ detection strategy, for example, on the orthogonality between codes used by different legitimate nodes during simultaneous transmissions. The transmission from node to is deemed successful at time , if , where is a fixed threshold. Let if , and zero otherwise. The sum in the denominator of the right hand expression in the above equation is referred to as the interference and ensures that the interference term in the denominator is finite almost surely. Since is exponentially distributed, multiple transmissions may be required for a packet to be successfully received at any node.

## 3 Main Results and Proofs

Our first objective is to show that with the power control policy described above, the expected time for a packet to be successfully received at the nearest neighbor in the destination cone is finite.

###### Definition 3.1

Let the minimum time (exit time) taken by any packet to be successfully transmitted from node to its nearest neighbor in the destination cone of the packet be

Let denote the Palm distribution of , conditioned to have a point at and let denote expectation with respect to . By an abuse of notation we will use and to denote and . This will cause no confusion since these are the only probabilities and expectations that are of interest.

###### Theorem 3.2

[Finite expected exit time] Suppose . Then for all sufficiently small, the space-time SINR graph with power control policy as described above satisfies .

###### Remark 3.3

The parameter controls per link data rate, larger the value of , larger is the per link data rate. The condition indicates that to support larger per-link data rate, one has to invest in getting a better (lower) interference suppression parameter, e.g. by lowering the chip rate in a wireless CDMA system. The condition also indicates that there is no free lunch, i.e., if one wants larger data rate and finite expected exit time, one has to have better interference suppression capability. An interesting upshot of the proposed power control policy is that the condition required for the theorem to hold is independent of the intensity of the PPP.

Proof: Without loss of generality, suppose that the origin , that is we will consider the point process under which, as we noted above will be denoted by . We tag a particular packet to be transmitted out of the node at , and suppose that the destination cone of this packet when it is at is . Denote the nearest neighbor of in by . We have dropped the time subscript from , since, as long as the packet is not successfully transmitted out of the destination cone remains the same. Let

(5) |

where

(6) |

Let if , and otherwise. Due to interference and the nature of the traffic arriving at the nodes, the choice of the destination cones are not independent across time slots at the same node as well as at different nodes. Hence to evaluate we need to condition appropriately. Let be the sigma field generated by the point process and the choice of cones made at all nodes of at times . Note that as long as the packet is not successfully transmitted out of , the transmission probability does not change. Let . Now

(7) |

Let be the event that the origin , and be the event that , but . Writing right hand side of (7) in terms of , and using the independence of the fading powers and the conditional independence of the transmission events, we get

(8) |

On the event , we have and

(9) |

for . (9) follows from (5) by using the fact that and taking expectation with respect to . This yields

(10) | |||||

where we have used the fact that . Let . We now find a lower bound for the expectation on the right hand side above that is independent of the choice of the cone. To this end observe that

(11) |

Suppose node transmits using cone in time slot . This fixes the transmission probability and power (where we have included the index to make the dependence on the cone explicit). Then

(12) | |||||

Let be the cone for which the right hand expression in (12) is minimized, i.e., node causes maximum interference at when its destination cone choice is . Let , denote the corresponding transmission probability and power respectively. Denote by an independent Bernoulli random variable with . The cone maximizes the interference contribution at due to transmission at . Define

(13) |

Substituting for in (10) along with the observation that given , we get

(14) |

Substituting from (14) in (8), we get

(15) |

where

The expected delay can then be written as

By the Cauchy-Schwartz inequality we get

(16) |

From the definition of the transmission probability , we get

(17) |

since the nearest neighbor distance in a cone has density

(18) |

It remains to show that

(19) |

(20) |

Taking expectations, first with respect to and then with respect to , we get

where follows by substituting for , and to obtain we have used the fact that the average power equals for and in particular and . Let . Substituting the above bound in (20) we get

(21) |

Note that since . Let denote a ball of radius centered at . Substituting the bound obtained in (21) in (19), we get

(22) |

where the last inequality follows by shifting the origin to and including points from an independent Poisson process of intensity , i.e., a PPP of intensity restricted to the set . Clearly is a PPP of intensity with the origin at . Hence by an application of the Campbell’s theorem in (22) and the fact that we get

(23) | |||||

since . This completes the proof of Theorem 3.2.

Next, we build upon Theorem 3.2, to show that the information velocity, that is, the rate at which packets flow towards their destination, is positive under the proposed power control mechanism. Information velocity is a key quantity in multi-hop routing. Larger the velocity, higher is the capacity of the network. The negative results in [13] on the infinite expected delay and zero information velocity are proved for delays that are averaged over the fading variables. In order to work with the delay variables directly and also to be able to use the ergodic theorem, we introduce several additional structures as we go along.

As a first step we track the movement of a tagged packet that starts at the origin and traverses the network as follows. Let be the time taken by this tagged packet starting at the origin to successfully reach its nearest neighbor in the destination cone . The packet is transmitted with power and the probability that it is transmitted in a time slot is . Going forward, if the packet is at node , , let be the time taken for the packet to successfully reach the nearest neighbor of in the destination cone . From the time the packet reaches to the time it is successfully delivered at , it is transmitted with power and transmission probability . One can think of the destination of the packet being located at and thus the destination cone for this packet is always a translation of the cone . We ignore the queuing and other delays at each node as is the standard practice. Note that the delays are not identically distributed. For instance, the point process as seen from the origin and from do not have the same distribution, since the area of ï¿½destination cone between and contains no other point of . In particular, is not a stationary sequence.

###### Definition 3.4

The information velocity of space-time SINR network is defined as

where is the distance of the tagged packet from the origin at time .

The following is the main result of this paper.

###### Theorem 3.5

Under the conditions of Theorem 3.2 and the proposed power control strategy, the information velocity , almost surely.

Proof: In order to prove this result, we first dominate the delays , by a stationary sequence , and show that a positive speed can be obtained even with these enhanced delays. This will be done by adding some additional points that will increase the interference and hence the delay. To this end, for all , let and , where . Note that the cones are non-overlapping since . Consequently, is a sequence of independent and identically distributed random vectors having the same distribution as the random vector , where and are independent with density of given by (18) and is uniformly distributed on

To nullify the effect of moving to the nearest neighbor, we progressively fill the voids with independent Poisson points as the packet traverses the network. This however leaves an increasing sequence of special points, the ’s in the wake of the tagged packet. The following construction is intended to take care of this issue and deliver a stationary sequence.

Let be a sequence of independent random vectors with each vector having the same distribution as . Define , recursively starting from so as to satisfy and . Observe that each , lies in the cone as shown in Fig. 3, and is a sequence of independent and identically distributed random variables.

Let be the delay experienced by the tagged packet in going from to when the interference is coming from the points of . For , let be an PPP of intensity independent of everything else. For , let be the delay experienced by the tagged particle in going from to when the interference is coming from the nodes in . Note that for the actual delay , that is, when the packet is at and trying to reach , the interference contribution is coming from the nodes in . To define , we have added additional interferers at . We assume that virtual interferers placed at behave similar to nodes of . Clearly , and furthermore the sequence is a stationary sequence. To prove the later assertion, consider any finite dimensional vector of delays . The distribution of this vector is a function of the distribution of the special points , the point processes and , which by our construction is translation invariant.

Suppose we showed that . Then by the Birkoff’s ergodic theorem [19], we have

almost surely, where is a random variable with mean .

Let be the counting process associated with an arrival process with inter-arrival times given by the sequence . Then the information velocity satisfies

The result now follows since is finite almost surely.

It remains to show that . The proof of this assertion proceeds along the same lines as the proof of Theorem 3.2 with in (6) replaced by , where

This would lead to a bound analogous to (16) with replaced by and replaced by , where is defined analogous to . By the conditional independence of and we get

Another application of the Cauchy-Schwartz inequality implies that the result follows if we show that

Proceeding as in (22)-(23), we get

since It remains to show that

(24) |

To compute the expression in (24) we proceed as we did in (20)-(21) and arrive at the following bound similar to the one in (22).

(25) | |||||

where and , . Let and note that . Let be a constant that will be chosen later. Define , and let That for all follows from (18). By the Chernoff bound, we have

It follows by the Borel-Cantelli lemma that, almost surely, there exists a such that for all . Hence for some constant ,

(26) | |||||

Using the fact that the function is non-increasing, we get

since by the comparison test. Since , as . So, we can and do choose such that . With this choice of , it follows from (26) that . This proves (24).

Conclusion

In this paper, we have proposed a new power control strategy and a non-ALOHA protocol to achieve finite expected time (delay) for a packet to successfully reach its nearest neighbor with the SINR model. In prior work, it is known that the expected time for a packet to leave its source is infinite with an ALOHA protocol, severely limiting the effectiveness of such wireless networks. The power control strategy chooses power to completely overcome the path-loss effect towards the nearest neighbor in a defined cone that contains the destination, and attempts transmissions with appropriate probability so as to satisfy an average power constraint at each transmitter. In addition to achieving finite expected delay, we also show that our policy achieves non-zero information velocity, that is defined as the ratio of the successfully covered distance and the delay needed to reach that distance, as time goes to infinity. As a result, packets can flow between any source-destination pair over multiple hops at a non-zero rate. Some outstanding questions that remain, are: what is the optimal choice of the angle of cone, larger the cone angle shorter is nearest neighbor distance and per-hop delay but requires more hops till the destination and vice versa, whats the best lower bound on the per-hop delay and the upper bound on the information velocity.

Acknowledgments

This paper has benefited from numerous useful discussions with Manjunath Krishnapur and D. Yogeshwaran.

## References

- [1] M. Haenggi, Stochastic geometry for wireless networks. Cambridge University Press, 2012.
- [2] P. Gupta and P. Kumar, “The capacity of wireless networks,” IEEE Trans. Inf. Theory, vol. 46, no. 2, pp. 388–404, Mar 2000.
- [3] S. Weber, X. Yang, J. Andrews, and G. de Veciana, “Transmission capacity of wireless ad hoc networks with outage constraints,” IEEE Trans. Inf. Theory, vol. 51, no. 12, pp. 4091–4102, Dec. 2005.
- [4] F. Baccelli, B. Blaszczyszyn, and P. Muhlethaler, “An Aloha protocol for multihop mobile wireless networks,” IEEE Trans. Inf. Theory, vol. 52, no. 2, pp. 421–436, Feb. 2006.
- [5] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to coverage and rate in cellular networks,” IEEE Trans. Commun., vol. 59, no. 11, pp. 3122–3134, 2011.
- [6] J. Andrews, S. Weber, M. Kountouris, and M. Haenggi, “Random access transport capacity,” IEEE Trans. Wireless Commun., vol. 9, no. 6, pp. 2101 –2111, June 2010.
- [7] R. Vaze, “Throughput-delay-reliability tradeoff with ARQ in wireless ad hoc networks,” IEEE Trans. Wireless Commun., vol. 10, no. 7, pp. 2142 –2149, Jul. 2011.
- [8] H. Kesten, “Aspects of first passage percolation,” in École d’Été de Probabilités de Saint Flour XIV-1984. Springer, 1986, pp. 125–264.
- [9] J. M. Hammersley and D. Welsh, “First-passage percolation, subadditive processes, stochastic networks, and generalized renewal theory,” in Bernoulli 1713, Bayes 1763, Laplace 1813. Springer, 1965, pp. 61–110.
- [10] R. Durrett, “Stochastic spatial models,” SIAM review, vol. 41, no. 4, pp. 677–718, 1999.
- [11] D. Mollison, “Spatial contact models for ecological and epidemic spread,” Journal of the Royal Statistical Society. Series B (Methodological), pp. 283–326, 1977.
- [12] ——, “Markovian contact processes,” Advances in applied probability, pp. 85–108, 1978.
- [13] F. Baccelli, B. Błaszczyszyn, and M.-O. Haji-Mirsadeghi, “Optimal paths on the space-time SINR random graph,” Advances in Applied Probability, vol. 43, no. 1, pp. 131–150, 2011.
- [14] P. Jacquet, B. Mans, P. Muhlethaler, and G. Rodolakis, “Opportunistic routing in wireless ad hoc networks: Upper bounds for the packet propagation speed,” IEEE J. Sel. Areas Commun., vol. 27, no. 7, pp. 1192–1202, 2009.
- [15] R. K. Ganti and M. Haenggi, “Bounds on the information propagation delay in interference-limited aloha networks,” in International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks, 2009. WiOPT 2009. IEEE, 2009, pp. 1–7.
- [16] D. Tse and P. Viswanath, Fundamentals of wireless communication. New York, NY, USA: Cambridge University Press, 2005.
- [17] Y. Yu, R. Govindan, and D. Estrin, “Geographical and energy aware routing: A recursive data dissemination protocol for wireless sensor networks,” Technical report ucla/csd-tr-01-0023, UCLA Computer Science Department, Tech. Rep., 2001.
- [18] S. A. Borbash, A. Ephremides, and M. J. McGlynn, “An asynchronous neighbor discovery algorithm for wireless sensor networks,” Ad Hoc Networks, vol. 5, no. 7, pp. 998–1016, 2007.
- [19] U. Krengel and A. Brunel, Ergodic theorems. Cambridge Univ Press, 1985, vol. 59.