A step forward performance optimization: Quasi Trace-

The insights about TO-STBC as a potential technique for rate improve- ment have been validated through the experimental results presented in the

previous paragraph. The transmission rate enhancement with respect to OSTBC is evident [52], while a high level of robustness to errors is main- tained. This fact suggests that a modest sacrifice in terms of reliability may bring significant benefits for rate, allowing in this context a further step towards the performance of SM. However, it is worth noting that the efficiency loss is attenuated by the nature of PPM scheme. In fact, PPM is known to be a bandwidth inefficient modulation providing lower bit rate in comparison with other schemes, but on the other hand it shows higher robustness to ISI. Taking into account this latter aspect, the adoption of a more relaxed condition than orthogonality or trace-orthogonality in STBC can lead to a rate improvement, ensuring reliability at the same time. In this context, it is possible to introduce a new property about the trace, defined as Quasi Trace-Orthogonal (QTO) condition, characterized by the following constraint to respect:

trace{ST_pSm} ≤ ξ, p 6= m, p, m ≤ LNT (5.20)

where the parameter ξ is:

0 ≤ ξ < blog2NTc (5.21)

The condition in eq. (5.2.4) states that the transmittable codewords are those ones retaining the property of having the trace less than ξ. As ξ de- pends on NT, eq. (5.2.4) represents a less strict constraint than eq. (5.2.1.2)

since for NT ≥ 4 the trace can assume values greater than 0 (for NT = 2, 3

eq. (5.2.4) and eq. (5.2.1.2) are essentially the same). This fact returns that the number of transmittable codewords in QTO-STBC is greater than in the case of TO-STBC, hence leading to a higher transmission rate. Table 5.3 reports the size of the codewords vocabulary for different PPM modulation orders and transmitters employed, and considering the four transmission schemes cited in the current chapter.

Table 5.3: Number of transmittable codewords as a function of scheme, transmitters and modulation order.

NT 2 3 4 L-PPM 2-PPM 4-PPM 8-PPM 16-PPM 2-PPM 4-PPM 8-PPM 16-PPM 2-PPM 4-PPM 8-PPM 16-PPM O-STBC 2 4 8 16 2 4 8 16 2 4 8 16 TO-STBC 2 4 8 16 2 4 8 16 2 4 8 16 QTO-STBC 2 4 8 16 2 4 8 16 2 16 55 256 SM 4 16 64 256 8 64 512 4096 16 256 4096 65536

For what concerns signal detection and codeword decoding, the proce- dures to be followed are the same discussed when introducing the receiver architecture for TO-STBC. In fact, the difference between TO and QTO relies only in the number of transmittable codewords.

The performance provided by the use of QTO-STBC have been evalu- ated both in terms of rate and BER, considering a 4x4 MIMO architecture

2 4 8 16 PPM Order 0 8 16 24 32 40 48 56 64 Rate (bits/frame) SM TO-STBC QTO-STBC

Figure 5.9: Rate comparison between TO-STBC, QTO-STBC and SM for different PPM constellations considering a 4x4 MIMO architecture.

(that is NT = 4) in order to highlight how the proposed technique is able to

outperform TO-STBC and approaching SM. The experiments consisted of a 4-PPM based data transmission, considering the same scenario as in Fig. 5.3 (only with fluid water surface). In this case, transmitter and receiver side have been equipped with an horizontal array of 4 SAM-1 modems and 4 hydrophones, respectively. Fig. 5.9 shows the achievable transmission rate of TO-STBC, QTO-STBC and SM, expressed in bits/frame according to the features of the SAM-1 modem. The rate is reported as a function of the modulation scheme, that is for 2-PPM, 4-PPM, 8-PPM and 16-PPM. By looking at the trends, it is possible to observe how QTO-STBC is in the middle of TO-STBC and SM. Regarding BER, the results obtained by the tests are collected in Table 5.4, and even in this case QTO-STBC is between TO-STBC and SM. Moreover, the following aspects can be underlined. The

4-PPM BER Scheme TO-STBC < 0.010 QTO-STBC 0.168 SM 0.218

Table 5.4: Communication BER for 4x4 MIMO system.

results in Table 5.4 and Table 5.2 refer to a 4x4 and 2x2 MIMO system re- spectively. Therefore, as in the first case interference and signals overlapping are stronger, it is reasonable that the BER values in Table 5.4 are higher than those in Table 5.2. By looking at Table 5.3 it can be noticed that in MIMO 2x2 4-PPM SM and in MIMO 4x4 QTO-STBC the number of trans- mittable codewords is 16 in both cases. Despite the same transmission rate, MIMO 4x4 QTO-STBC outperforms MIMO 2x2 SM in terms of BER, hence

demonstrating the higher reliability provided by QTO-STBC with respect to SM. Furthermore, as expected, the BER measured for TO-STBC is lower than the BER for QTO-STBC, but this fact is paid in terms of rate since the number of transmittable codewords in MIMO 4x4 4-PPM TO-STBC is lower and equal to 4. Finally, it can be observed that for TO-STBC the number of used codewords is the same both in MIMO 2x2 and MIMO 4x4, so the resulting rate is the same but also the measured BER is comparable. Summarizing, it is worth remarking that for the tests described above equalization has not been considered. Therefore, by resorting to a well- performing equalization scheme may lead the measured values of BER to be all rescaled downward.

Error control and access

Although the system architecture at the physical layer is designed in order to provide performance as good as possible, it is anyway unlikely that the resulting communication is completely error free. This problem is also han- dled at the data link layer, where the cooperation among network nodes is exploited for implementing error control strategies making data transmis- sion more reliable. In this context, especially in the underwater scenario, it is also important to understand which is the most suitable portion of spec- trum where the transmission may take place, so that the interference with other communications or the impairments caused by the nature of the chan- nel are minimized. Therefore a mechanism for spectrum sensing and access results to be fundamental to achieve high efficiency. Recently, the NATO Centre for Maritime Research and Experimentation (CMRE) released the standard JANUS proposing a multiple-access acoustic protocol designed to be compatible with different devices and to be employed for both military and civilian purposes [53].

Following this direction, in the current chapter presents an ARQ-based (Automatic Repeat on reQuest) strategy for error detection and correction, while a restatement of the cognitive paradigm is proposed for managing the spectrum access in underwater acoustic networks.

6.1

ARQ protocols for error control

Bi-directional underwater acoustic communications usually request the trans- mission of some control information among nodes in order to govern the data flow. Attenuation, multipath and noise affecting the signal propagation may lead to errors during the data detection phase, weakening the integrity of the received information. In this regard, the effects induced by the underwater acoustic channel can be, sometimes, partially counterbalanced with the use of coding, that is Forward Error Correction (FEC) and Automatic Repeat reQuest (ARQ), but in some cases ARQ is more reliable than FEC especially

in the presence of long error bursts [54]. However, both approaches consider the introduction of overhead information for different purposes. ARQ uses overhead in order to perform error detection and FEC uses overhead for er- ror correction. Moreover, it is worth noting that FEC codes are not always able to correct errors, so it could happen that the received information is anyway erroneous or incomplete. When instead using ARQ, the integrity of the data is guaranteed by the fact that each frame is retransmitted until it is correctly received.

ARQ is usually implemented in a point-to-point link according to three different schemes that are Stop-and-Wait (S&W), Go-Back-m (GBm) and Selective Repeat (SR). When using S&W, the transmitter sends a frame and waits for a positive acknowledgement (ACK) by the receiver indicating right frame detection, before transmitting the next one. If no ACK message is received within a pre-defined time window, the packet is retransmitted. When GBm is considered, the transmitter can send a sequence of frames, up to M , while waiting for ACKs. The receiver replies with an ACK, reporting also the frame number, that indicates that all the frames until the current one have been detected without errors, so if the ACK referring to the m-th frame is not received the broadcaster node will retransmit frames from the mth to the last one sent. In case of SR, the receiver, by using different ACKs and negative ACKs (NACKs), notifies the transmitter which are the corrupted frames that need to be retransmitted.

ARQ techniques are not new in underwater acoustics. In fact, in [55] a particular ARQ scheme has been implemented by exploiting the long prop- agation delay in a multi-hop acoustic channel. By choosing the packet size so that transmission time remains smaller with respect to propagation de- lay it is possible to establish concurrent bidirectional transmissions between nodes, and the backward overhearing is used as a somewhat implicit ACK. In this way overhead and transmission latency are reduced, and enhance- ment in power saving is also obtained.

A strategy for continuous ARQ is described in [56]. Even in that case, by considering the underwater propagation delay, a transmission scheme based on multiple data packets and ACKs cross-like manner dispatch has been proposed. Simulations results show that the throughput can be improved with respect to the simple S&W.

Furthermore, in [57], two Hybrid ARQ schemes have been presented in order to tackle the problem of error correction capability decrease at low Signal-to- Noise Ratio (SNR). By studying the impact of underwater acoustic channel features on the communication, a particular error control mechanism has been proposed. That strategy allows an improvement in terms of reliability, if compared to simple S&W. The obtained results have been then exploited in [58] to analyze the performance of the proposed schemes in a multi-user UWAN scenario.

introduced in [59]. The transmission of the first frame is operated according to the S&W mode so allowing, through the ACK, to measure the round trip time (RTT). If RTT is sufficiently long, the scheme then switches into a SR mode where multiple frames can be sent before receiving an ACK.

In [60] the analysis of the underwater acoustic channel and the variability of its parameters has led to the design of a variable window GBm scheme. With this approach the time window for feedback signaling is dimensioned according to the length of the variable signal propagation delay. This allows to achieve a higher efficiency in terms of channel utilization and error cor- rection than that one given by the simple GBm scheme.

Last, an alternative to ARQ is instead given by [61]. Due to long propaga- tion delays, the feedback waiting time could be very long, so communication through underwater acoustic channels becomes inefficient. The system per- forms a coded data packets transmission, while feedback signal is used by the receiver to communicate information about channel variation to the trans- mitter, so that it can adjust the transmit power and packet size.

All the above mentioned works from the literature tackle the issue of optimizing ARQ schemes without directly consider the physical layer issues, and, most of them, focus on point-to-point links. The contributions that deal with multiple nodes consider them as relays in a multi-hop scenario. Moreover, the analyses have been carried out only by means of simulations, so there are no data coming from field experiments.

Basing on the above considerations, the following section presents the adaptation of some typical rules of the three main ARQ strategies, Stop&Wait, Go-Back-m and Selective Repeat respectively, by considering only the trans- mission of an (explicit) NACK whenever bit errors are detected in a received frame. In this regard, the NACK message is represented by a single pulse sent on a sub-bandwidth whose center frequency depends on the index of the error-affected frame. Hence the NACK spectrum does not depend on the node that transmits it, thereby avoiding the need to implement multi-user detection techniques. The reference communication scenario is a broadcast acoustic sensor network, very often used in underwater applications (but largely unexplored in the literature) such as data collection activities where several sensors placed underwater are controlled by a master node that has usually to (i) communicate instructions and parameters about the measur- ing activity to the sensors, (ii) carry out the firmware update of the devices remotely (so without removing them from their place), (iii) check the pres- ence and safety of potential divers by sending periodically some broadcast alive messages. In particular the structure presented is characterized by one transmitter and multiple receivers that employ separated bands for feedfor- ward transmission and feedback signaling. A frequency domain mechanism for NACKs detection is employed. Using the proposed approach grants sev- eral benefits with respect to the standard ARQ protocols and FEC that are

summarized in Table 6.1.

Table 6.1: Proposed feedback scheme compared with standard ARQ proto- cols and FEC

STANDARD ARQ PROPOSED FORWARD ERROR

PROTOCOLS MECHANISM CORRECTION

X Received data integrity is guaranteed % FEC codes could not be able_{to correct all the errors} % Communication throughput reduced due to

feedback messaging and frame retransmission

% Too much feedback signaling X Explicit-NACK/Implicit-ACK use

(an ACK/NACK for each user) saturate allows signaling reduction if X No feedback signaling the transmission resources compared to explicit ACK-based schemes and no frame retransmission

% Possible errors X NACK recognition performed by means in ACK and NACK detection of a simple energy detector

% Multi-user detection is needed X Due to the network structure, X No multi-user detection for colliding signals no multi-user detection is needed is requested

X The index of a corrupted frame % Frame numbering necessary equals to that of the sub-bandwidth where

to avoid ambiguity the corresponding NACK is sent over

% Large error correction control bits X Few error detection control bits are requested for error detection

and correction % High system complexity X Low system complexity and latency