WIRELESS Sensor Networks (WSN) have drawn the attention of the research community, driven by a wealth of theoretical and practical challenges. Significant results in this area have ushered in a surge of civil and military applications. As of today, most deployed wireless sensor networks measure scalar physical phenomena like temperature, pressure, humidity, or location of objects. In general, the applications they are designed for have low bandwidth demands, and are usually delay tolerant. More recently, the availability of inexpensive hardware such as CMOS cameras and microphones that can ubiquitously capture multimedia content from the environment has fostered the development of Wireless Multimedia Sensor Networks (WMSNs),
i.e., networks of wirelessly interconnected devices that can retrieve video and audio streams, still images, and scalar sensor data. By enabling new applications such as multimedia surveillance, traffic enforcement and control systems, advanced health care delivery, structural health monitoring, and industrial process control, WMSNs will be a crucial component of mission-critical networks to protect the operation of strategic national infrastructure, provide support for emergency and crisis intervention, and enhance infrastructure for tactical military operations.
Many of the applications described above require the sensor network paradigm to be re-thought in view of the need to deliver multimedia content with predefined levels of quality of service (QoS). QoS-compliant delivery of multimedia content in sensor networks is a challenging, and largely unexplored
task. First, embedded sensors are constrained in terms of battery, memory, processing capability, and achievable data rate , while delivery of multimedia flows may be a resource intensive task. Secondly, in multi-hop wireless networks the attainable capacity of each wireless link depends on the interference level perceived at the receiver. Hence, capacity and delay attainable at each link are location dependent,
vary continuously, and may be bursty in nature, thus making QoS provisioning a challenging task. Lastly, functionalities handled at different layers of the communication stack are inherently and strictly coupled due to the shared nature of the communication channel. Hence, different functionalities aimed at QoS provisioning should not be treated separately when efficient solutions are sought, i.e., a cross-layer design approach is needed.
Existing sensor networks are mostly based on variants of the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) Medium Access Control (MAC) protocol. CSMA/CA has demonstrated to be an effective mechanism to distributively share a common wireless channel among uncoordinated devices. However, it requires mutually exclusive transmissions, i.e., when a device is receiving data, transmissions from all the devices in its transmission range are impeded. Mutual exclusion is achieved by distributively coordinating the transmissions of different sensors mainly by means of two mechanisms, i.e., carrier sense and random timers to defer transmissions. While random timers lead to variable and uncontrollable access delays, carrier sense causes considerable energy consumption for idle listening ; still, frequent collisions occur due for example to the well-studied hidden node problem, in turn leading to increased energy consumption and delays. The transmitted power of currently off- the-shelf motes, such as Crossbow's MicaZ, based on the Chip on 2420 chipset, is still high, in the order of 1mW. Exact Tx: Rx: Idle power ratios depend on hardware but idle power is in general not negligible and accounts for a considerable portion of the overall energy consumption.
Introducing sleep periods to reduce idle listening reduces the energy consumption at the expense of latency and coordination complexity. For the reasons above, although recent proposals have modified existing protocols based on CSMA/CA and geographical routing to provide delay-sensitive and error-resilient services in sensor networks, the application requirements of WMSNs call for a new design perspective and next-generation wireless technologies. Hence, in this paper we propose new cross-layer communication architecture to reliably and flexibly deliver QoS to heterogeneous applications in WMSNs, by leveraging and controlling interactions among functionalities handled at different layers according to applications requirements. Our design is based on the MIMO transmission technique. The MIMO cross layer technology has the potential to enable low power consumption, high data rate communications within
tens of meters, which make it an ideal choice for WMSNs. The MIMO cross layer is particularly appealing for WMSNs as it enables high data rate, very low power wireless communications, on simple-design, low-cost radios . Its fine delay resolution properties are well-suited for dense multipath environment. Importantly, interference mitigation techniques allow realizing MAC protocols that do not require mutual temporal exclusion between different transmitters. Finally, the large instantaneous bandwidth enables fine time resolution for accurate position estimation and network synchronization, while at the same time the low power spectral density enables military covert operations.
The remainder of the paper is organized as follows. In Section II, we discuss previous work on multi-hop networking with UWB. In Section III we outline the main design principles, and describe the proposed cross-layer architecture. In Section IV, we introduce the considered system model. In Section V, we describe the routing and admission control functionalities. In Section VI we describe the medium access control and the proposed dynamic code assignment and scheduling policies. In Section VII we discuss performance evaluation results while in Section VIII we draw the main conclusions.
II. RELATED WORK
There is a vast literature on physical layer aspects of the UWB technology. Excellent comprehensive surveys of the UWB transmission technique, and of localization techniques for UWB systems, are provided in the past. Although, like CDMA, TH-IR-UWB is a multi-user radio technology, non-zero cross-correlation between time hopping sequences, time-asynchronicity between sources and the strong effect of multipath propagation require for suitable MAC and higher layer solutions. However, higher layer solutions for multi-hop wireless networking with UWB are not mature yet. In one paper, Cuomo et al. investigate the problem of joint rate and power assignment for TH-IR-UWB, and formulate it as an optimization problem. They show that when the objective is to maximize the aggregate data rate, the optimal solution always corresponds to points where individual devices transmit at the maximum power, or do not transmit at all. The finding is confirmed in one paper , where the authors show that power control is not required and may even be suboptimal for wireless networks in the linear regime, i.e., when the achievable data rate is linearly dependent on the signal-to interference- plus-noise ratio (SINR) at the receiver. Note that this is a peculiar characteristic of TH-IR-UWB, and does not hold in general for relatively narrowband systems such as CDMA or IEEE 802.11. The result holds both when the objective is to maximize the data rate under power constraints and when the objective is to minimize the power consumption under constraints on minimum data rates.
Based on the above finding, an uncoordinated MAC protocol for low-power UWB devices is proposed .While most existing protocols manage interference and multiple access through power control or mutual exclusion, the MAC proposed in which is based on rate control, i.e., it dynamically adapts the channel code based on the interference at the receiver. The proposed design takes advantage of the nature of pulsed TH-UWB to further propose an interference mitigation scheme that alleviates the need for an exclusion scheme. Each device is always allowed to transmit and continuously adapts its channel code to the interference experienced at the destination. Such MAC layer does not need coordination among neighbors that are not involved in the communication, and is shown by simulation to achieve a significant increase in network throughput compared to alternative designs. In one paper, a centralized MAC protocol designed to provide QoS support for multimedia traffic in UWB-based wireless local area networks is proposed. Unlike our work, none of the previously proposed solutions consider the problem of satisfying and differentiating between QoS requirements of the overlying applications. Moreover, no existing practical solution considers the cross-layer interactions between routing, MAC and physical layer functionalities.
III. DESIGN PRINCIPLES AND CROSS-LAYER
CONTROLLER
In this section, we overview the principles that guide our system design. We assess the benefits of our design in view of the performance objectives and of the characteristics of WMSNs, and describe the cross-layer control architecture of the MIMO sensor.
Network Layer QoS Support enforced by a cross-layer controller.
The proposed system provides QoS support at the network layer, i.e., it provides packet-level service differentiation in terms of throughput, end-to-end packet error rate, and delay. The architecture of the proposed controller is shown in Fig. 1. The cross-layer control unit (XLCU) configures and controls the networking functionalities at the physical, MAC, and network layer, based on a unified logic that takes decisions based on
i) Application requirements specified by the application layer;
ii) The status of the functional blocks implementing
the networking functionalities. In this way, cross-layer interactions can be leveraged without sacrificing on upgradeability, modularity, and ease of system design.
Geographical Forwarding.
Time-based localization techniques in UWB allow ranging accuracy in the order of centimeters [12]. Hence, our module leverages geographical information to provide QoS, as further explained in Section V. Positioning capabilities are needed in sensor networks to associate physical meaning to the
information gathered by sensors. Moreover, knowledge of the position of each network device allows for scalable routing solutions [26].
Hop-by-Hop QoS contracts. End-to-end QoS requirements are enforced through local interactions. Each device is responsible for locally guaranteeing given performance objectives. The global, end-to-end requirement is thus guaranteed by the joint local decisions of the participating devices, as further explained in Section V.
Receiver-centric scheduling for QoS Traffic.
In multihop wireless environments interference is location dependent For this reason, we provide QoS through receiver-centric scheduling. The receiver can be responsive to the dynamics of the channel based on local measurements and consequently control loss recovery and rate adaptation, thus avoiding feedback overheads and latency.
MIMO Physical/MAC layer. We rely on an integrated MAC and physical layer based on UWB. Like CDMA, MIMO allows multiple transmissions in parallel. Conversely, typical MAC protocols for sensor networks, such as those based on CSMA/CA, require mutual temporal exclusion between neighboring transmitters. This allows devising MAC protocols with minimal coordination, as will be discussed in Section VI. In spite of the recent advances in the design of low-complexity transmitters and receivers, the hardware complexity of CDMA transceivers is still relatively high. Instead, MIMO transceivers are simple to realize.
Dynamic Channel Coding.
As previously discussed, power control is not beneficial in TH-IR-UWB. Hence, adaptation to interference at the receiver is achieved through dynamic channel coding, which can be seen as an alternative form of power control, as it modulates the
energy per bit according to the interference perceived at the receiver [11]. This will be explained in Section VI.
IV. SYSTEM MODEL
A. Network Model
The sensor network is represented as a graph G(V, E),
where V = {v1, ..., vN} is a finite set of devices (nodes) in a finite-dimension terrain, with N = |V|, and E is the set of links among nodes, i.e., eij ∈ E iff nodes vi and vj are within each other's transmission range. Node vN (also N for simplicity) represents the sink.
B. Physical Layer Model
MIMO cross layer transmits sub nanosecond pulses (in the order of hundreds of picoseconds), referred to as monocycles. We model a monocycle as the second derivative of a Gaussian pulse. Time is slotted in chips of duration Tc, and chips are organized in frames of duration Tf = NhTc, where Nh is the number of chips per frame. Each user transmits one pulse in one chip per frame, and determines in which chip to transmit based on a pseudo-random time hopping sequence (THS). The train of monocycles is modulated based on pulse position modulation (PPM), i.e., a '1' symbol is carried by a monocycle delayed by a time δ with respect to the beginning of the chip, while a '0' symbol begins with the chip. In the above model, the signal s(k)(t, i) generated by the kth user to convey the ith symbol is expressed.
sequence of the kth source, with
0 ≤ c(k)j≤ Nh − 1,{d(k)i} is the information-bearing sequence, d(k)I ∈ 0, 1 , Eb
represents the energy per bit and Ns the number of pulses to represent a single bit. Clearly, by increasing the number of pulses per bit Ns one can increase the robustness to multiuser interference, at the expense of the data rate, which is expressed as R = 1/NsTf . This technique is referred to as pulse repetition coding. Each transmitter i transmits at a specified raw pulse rate R0,i = 1/Tf,i. Assuming that pulses generated at the physical layer have a width Tp, we transmit at a peak power Ppeak = Epeak/Tp = 0.28mW, i.e., the limits allowed by regulations and hardware constraints . Given a frame of 277 chips, this corresponds to an average radiated power of about 1 μW, considerably lower than what radiated by state-of-the-art motes.
C. Multi-path Channel Model
We model the channel according to the IEEE 802.15.4a standardization group model [27]. The model, specifically developed for sensor network applications, is based on extensive measurements of UWB channels and can be parameterized for indoor residential, indoor office, outdoor, and industrial
environments amongst others. The path loss exponent α depends on whether there is line of sight between the transmitter and the receiver or not, on the antenna gain and efficiency. Note that shadowing can be neglected in 802.15.4a simulations. The impulse response of the channel is modeled according to a modified Saleh-Valenzuela model .The model reproduces the clustering phenomenon observed in several MIMO measurements, and accordingly assumes that multipath components arrive in clusters, and that there is independent fading for each cluster and for each ray within the cluster. The interarrival times between two consecutive rays in a cluster and between two consecutive clusters are negative exponentially distributed with parameters λ and Λ, respectively. The power delay profile (PDP), i.e., the mean power of the different paths, is exponential within each cluster.
D. Coding
The proposed system includes a channel encoder block that encodes raw data bits into encoded bits that are then transmitted as pulses by the MIMO modulator. The channel encoder adds redundancy to combat channel impairments and multi-user interference. As discussed in more detail later, our proposed system leverages dynamic channel coding to adapt the transmission rate to the interference perceived at the receiver. The encoder at node i receives a block of L uncoded bits, selects the encoding rate RE,i, which represents the number of data bits per encoded bit. The set of available codes RE depends on the chosen family of codes C. Different families of codes, such as pulse repetition codes or rate compatible punctured codes, have different performance and different levels of complexity.
V. DISTRIBUTED ADMISSION CONTROL FUNCTIONALITY
The proposed system is based on the concept of Hopby- Hop QoS contracts. Each device in the end-to-end path is responsible for locally guaranteeing given performance objectives to devices that are obtaining a service from it. The global, end-to-end requirement is thus guaranteed by the joint local interactions of the participating devices.Let us consider a flow ψa(δa, βa, ζa) generated at node i that requires service. A multi-hop path from i to the destination N needs to be established, with maximum end to-end delay δa, minimum guaranteed bandwidth βa, and maximum end-to-end packet error rate ζa.
The required bandwidth βa needs to be provided at each hop. As far as delay and packet error rate are concerned, given a potential next hop j, on link eij we can allow a delay δij proportional to the geographical advance of the packet towards the destination at that hop. For example, if the first hop towards the destination guarantees an advance that equals one third of the total geographical distance towards the destination, then one third of the total allowed end-to-end delay can be allowed to that hop. A similar procedure is used to derive the allowable packet error rate on a single hop.
VI. MEDIUM ACCESS CONTROL, SCHEDULING AND RATE ASSIGNMENT
In this section, we discuss how our cross-layer module achieves coordination to share the transmission medium among devices, schedules transmissions of data packets and assigns data rates to different flows based on the application requirements.
B. Receiver-centric Scheduling
For unicast transmissions, a pseudo-random time hopping sequence THS(j) is generated using the identity of the receiver j as the seed of the random number generator, while for multicast transmissions the time hopping sequence THS(i) is generated based on the identity of the transmitter i. Coordination of medium access is still needed to:
1) Prevent collisions at the receiver. When a device
i is receiving data from a device j, no other device should transmit data intended for i (i.e., using THS(i))
simultaneously, as we assume that i is endowed with a simple single-user receiver.
2) Avoid idle listening. Each device should be tuned to the wireless channel only when incoming transmissions for itself are occurring, i.e., each device should consume energy only when actually receiving data.
3) Avoid wasteful transmissions. When a device i is
transmitting data to j, j's receiver must be tuned to THS(j) to listen for incoming transmissions.Our objective is therefore to devise a medium sharing policy that achieves the above objectives with simple coordination. Each device is responsible for scheduling transmissions of data packets from its upstream nodes, i.e., the devices it is offering a service to, i.e., ∀u ∈ Fi. Device i prepare a SCHEDULE packet, that is transmitted at periodic intervals Δs. The scheduling period Δs is known to all network devices. The phase Φi s is communicated by i to its upstream nodes in the CONTRACT_ESTABLISHED message. The SCHEDULE packet is broadcast by i and all its upstream nodes receive it by periodically tuning their MIMO receiver to THS(i). Hence, conflict-free scheduling can be achieved in a very simple way. This is only paid in terms of flexibility, as all incoming flows have to be transmitted downstream through the same next-hop, i.e., multi-path routing does not fit in this framework. However, this is a price worth paying for the simplicity achieved. We determine the actual scheduling of packets from upstream nodes based on a procedure inspired by the wireless fair scheduling (WFS) paradigm. WFS] is a family of solutions designed to guarantee delay-bounded and throughput guaranteed access in single-hop, single-rate wireless packet networks (i.e., cellular networks). Most of these solutions are based on wireless adaptations of the packetized version of
the Generalized Processor Sharing (GPS) paradigm. We consider a wireless fair service approach [28] and extend it for a MIMO multi-rate, multi-hop environment.
VII. PERFORMANCE EVALUATION
To assess the performance of the proposed solution, we have developed two software simulation tools, i.e., a bit-level physical layer simulator of the MIMO communication architecture in Matlab, and a discrete-event object-oriented packet-level simulator in Java. The physical layer simulator models generation, modulation and coding of Gaussian monocycles, convolution with the multi-path-affected MIMO channel, interference from concurrent transmitters, and reception with a correlation receiver affected by multi-user interference and noise.
Bit Error Rate with increasing number of users, for different Pulse Repetition Codes, for SNR=30dB, no multipath (a) for SNR=30dB,no multipath (b) for SNR=30dB, with multipath
VIII. CONCLUSIONS
We have described the design of cross-layer communication architecture to provide QoS in wireless multimedia sensor networks based on MIMO communications. The architecture is based on an innovative design that aims at providing differentiation in the domains of throughput, delay, reliability, based on a modular cross-layer controller that performs admission control, routing, scheduling, bandwidth assignment and coding to satisfy application requirements. Performance evaluation shows that the architecture is a promising solution to satisfy the performance targets of WMSNs. In particular, delays are very low and with low jitter, and throughput is fairly constant in time.
