Carrier Sense Random Packet CDMA Protocol in Dual-Channel Networks

Code resource wastage is caused by the reason that many hopping frequency (FH) sequences are unused, which occurs under the condition that the number of the actual subnets needed for the tactical network is far smaller than the networking capacity of code division networking. Dual-channel network (DCN), consisting of one single control channel and multiple data channels, can solve the code resource wastage effectively. To improve the anti-jamming capability of the control channel of DCN, code division multiple access (CDMA) technology was introduced, and a carrier sense random packet (CSRP) CDMA protocol based on random packet CDMA (RP-CDMA) was proposed. In CSRP-CDMA, we provide a carrier sensing random packet mechanism and a packet-segment acknowledgement policy. Furthermore, an analytical model was developed to evaluate the performance of CSRP-CDMA networks. In this model, the impacts of multi-access interference from both inter-clusters and intra-clusters were analyzed, and the mathematical expressions of packet transmission success probability, normalized network throughput and signal interference to noise ratio, were also derived. Analytical and simulation results demonstrate that the normalized network throughput of CSRP-CDMA outperforms traditional RP-CDMA by 10%, which can guarantee the resource utilization efficiency of the control channel in DCNs.


Introduction
Wireless ad hoc networks that are widely deployed for tactical communication are mainly constructed by the de-synchronization or synchronization methods.Compared with a de-synchronized network, a synchronized network has many advantages in network interoperability, large networking scale, strong track interference resistance and high throughput.Hence the synchronized network is mainly used in a large-scale tactical environment.On the battlefield, a common tactical network should be divided into a fewer subnets for tactical communication.Since the number of subnets needed in a common tactical network is far smaller than the theoretical networking capacity, a large number of frequency hopping (FH) sequences are unused, which causes serious code resource wastage.Theoretical networking capacity denotes the maximum number of subnets that the network can be divided into, under the condition that the network is orthogonal and synchronized.The theoretical networking capacity is equal to the number of frequency points used in the network.To reduce code resource wastage, dual-channel network (DCN), which consists of one control channel and multiple data channels and employs a hierarchically distributed ad hoc networks based on a fully connected network, was proposed [1].The network reliability of DCN depends on the anti-jamming capability of the control channel, which is one of the key technologies of DCN design.As a result of advanced channel access and anti-jamming capability, the spread-spectrum ad hoc network technology is introduced into the design of the control channel to enhance the interference resistance [2].
We assume that spread spectrum ad hoc networks is slotted [3], but we do not focus on synchronization.In this type of networks, the nodes are always divided into multiple fully connected networks, which are distinguished from each other through multiple spreading codes [4].Because of direct-sequence spread spectrum signaling, the receiver in a typical network can distinguish the concurrently received packets overlaying in both the space and time domains.The transmitter or receiver should know the spreading code used for the reference packet.The rule of how to assign a spreading code is the purpose of code assignment.Four categories of spreading-code assignment protocols have been proposed [5][6][7][8][9]: Common-code; Receiver-based; Transmitter-based; Hybrid [5].
Random packet code division multiple access (RP-CDMA) [10], which is a packet-level spread-spectrum multi-access scheme, provides private, connectionless and uncoordinated random channel access.RP-CDMA also provides a novel code assignment protocol, which is different from the traditional protocols described above.Its packet format consists of a short packet header and a long packet payload.Each header is spread with a common code identical for all nodes, while the payload of each packet is spread with a randomly generated spreading code.Each receiver consists of a header detector and a payload detector.RP-CDMA has been originally applied separately to a centralized station network and a satellite network based on the Aloha system.In these two networks, the receiver always transmits acknowledge packets in another channel because of bi-directional links, which does not cause any collision between the data packet and acknowledge packet [11].In addition, Kempter et al. applied RP-CDMA to ad hoc networks with simple and flat topology, and they focused on the multiuser detector performance of RP-CDMA, but did not design any MAC protocol that considered the unidirectional link of half-duplexed terminals.T. Mortimer et al. [12] designed a MAC protocol for an ad hoc network with RP-CDMA, but they did not consider the clustered topology or provide the analytical models of spread spectrum ad hoc networks with RP-CDMA.
In this paper, we propose a new MAC protocol called carrier sense random packet CDMA (CSRP-CDMA) to improve the interference resistance and the resource utilization efficiency of the control channel.We provide performance evaluation models for slotted spread-spectrum ad hoc networks by considering intra-cluster and inter-cluster multi-access interferences (MAIs), Additional White Gaussian Noise (AWGN), and packet collisions.Moreover, the performance measures such as packet transmission success probability, network throughput, and the signal interference to noise ratio (SINR) are investigated.Based on the presented models, the effects of spreading gain, network scale, detection capability, and signal propagation distance on network performance are analyzed.Finally, we simulate the proposed protocol in comparison with the existing protocols.The analytical and simulation results demonstrate that, compared to traditional RP-CDMA, the normalized network throughput based on CSRP-CDMA can reach 0.48, which guarantees the resource utilization efficiency of the control channel.The results can be used as the optimal parameter configuration for the design of the control channel.
The rest of the paper is organized as follows.Section 2 describes DCN.The system model is provided in Sec. 3. Section 4 presents CSRP-CDMA protocol.The performance evaluation model is presented in Sec. 5.The analytical and simulation results are discussed in Sec. 6.Finally, Section 7 concludes the paper.

Dual-Channel Networks
Code resource wastage is often caused in the traditional tactical networks by using the method of synchronization networking.Thus, with the orthogonal and synchronized networks as an example (Fig. 1), the group net uses the frequency division networking method to distinguish from each other.The frequency set in group net no. 1 is {f 0 , f 1 , f 2 , …,f n -1 }, where n is the number of frequency points.The theoretical networking capacity is n under the condition that the network is orthogonal and synchronized.As shown in Fig. 1, there are m sequences (Seq.1 Seq.m) assigned to m subnets in the group net no. 1, and the other n -m sequences (Seq.m + 1 Seq.n) are unused, which means the frequency points corresponding with these unused sequences are code resource wastage because no radio can use them.Hence, code resource wastage is very severe when m is very small.DCN is proposed to improve resource utilization efficiency.It uses a hierarchically distributed network structure and its sub-cluster is a fully connected network, which implies all the nodes in a sub-cluster can communicate with each other in one hop distance.The dual-channel structure of a node consists of a control channel and multiple data channels.The control channel is assigned before networking with the same FH sequence to exchange the network management and route control messages, and this FH sequence is fixed until the whole network is dismissed.The data channel will be established when a node pair has the request of data transmission.To transmit data packets, all nodes in the data channel will be assigned with the same idle FH sequence chosen from the idle code resource pool.In brief, each FH sequence is deemed as a frequency-pointvarying channel, and the DCN must contain multiple frequency points during each FH slot period.Since a DCN has one single transceiver, its node stays at either control channel or data channel by switching the single transceiver via a time division mechanism.The DCN has the following main merits: 1) the idle code resource can be reused dynamically by any subnet building multiple data channels between communication radio pairs; 2) the inter-cluster communication of traditional networks is realized with the use of extra convertor equipment, but that of DCN can be realized by building a data channel with a special common FH sequence only for the communication radio pairs.Fig. 2 illustrates the structure of DCN.The solid lines denote the data channel, and the dotted lines denote the control channel.In addition, the virtual backbone network consists of all the gate nodes and subnet-header nodes in the DCN for inter-cluster communication.Assuming the  node mobility in the DCN is low, we do not focus on high mobility in this paper.For simplicity, the subnet is called "cluster" instead of "sub-cluster" in the rest of the paper.

System Model
In this section, we describe the details of a DCN sys tem model composed of node distribution model, packet format, signal transmission model and receiver model.The assumptions of the system model are as follows: 1) the whole network is based on slot synchronization; 2) the whole network deploys a hierarchically distributed network structure and each cluster is based on a fully connected network (Fig. 2); 3) packet transmission can be interfered with the collisions, MAI [13] and channel noise.The major notations used in the receiver model are listed in Tab. 1.

Node Distribution Model and Packet Format
For simplicity, we assume that all nodes are deployed at the designated area in a Poisson random distribution, regardless of the tactical task requirement.During one packet duration, the distribution of sending-node number in Area A follows a Poisson process at rate  s .Thus, the prob- ability P s that the sending-node number k in any cluster during one packet duration is computed as follows: Fig. 3 shows the packet format used for the system.On the assumption that a packet header is only allowed to be sent or received during only one slot duration that is defined as the minimum time unit used for this system.Thus, we set the header duration T h to one slot.Hence, the duration of the packet payload satisfies Thus, the packet length

Signal Transmission Model
We denote the distance between a sending node and a receiving node as d, the transmitted power of the sending node as Power t , and the received power of the receiving node as Power r .Thus, Power r . is expressed as [14]: where r is the boundary between the near field and the far field, n is the attenuation factor, and C is a constant.

Receiver Model
Fig. 4 illustrates the receiver structure used for the system.The receiver structure consists of a packet header detector and a packet payload detector.The packet header detector is a single-user matched filter (MF), while the packet payload detector is a multiuser detector such as MF [15], Decorrelator [16], Minimum Mean Square Error (MMSE) [17], and Successive Interference Cancellation (SIC) [18].In our work, we use a single-user MF as the packet header detector and a MMSE as the packet payload detector.When multiple packet payloads coming from different source nodes arrive at the receiver simultaneously, they can be recovered successfully by the packet payload detector using multiple payload spreading codes.Because the packet header and the packet payload use different spreading codes, the detection of the packet header is not interfered with the packet payload.However, the detection of packet payload depends on the multiuser detection capability when there is too much interference from other packet payloads.Therefore, the multiuser detection capability constrains the system's anti-MAI performance.Furthermore, the packet header collision, which is caused by multiple headers arriving at the receiver during the same slot duration, is the only potential packet collision, because packet headers in a cluster use the same spreading code.
Generally, since the ratio of header length to payload length is very small, the channel load of the packet header is extremely slight which means the probability of header collision is very low.We assume that the packet header detector uses a single-user matched filter, then SINR ( ) mf j  of packet j at the output of header detector is calculated as where  2 is the normalized variance of AWGN.Then packet j will be detected successfully if ( ) .
Then we assume that the packet payload detector is an MMSE multiuser detector and that the threshold of MMSE is mmse  .Similarly, the payload of packet j will be detected successfully when the SINR

CSRP-CDMA Protocol
This section provides the details of CSRP-CDMA.Based on a reported MAC protocol [12], CSRP-CDMA is mainly added with a packet-segment-based acknowledgement policy (PSAP) and carrier sensing random packet mechanism (CSRPM).For simplicity, we neglect the other parts of CSRP-CDMA, which are similar to RP-CDMA [12].Thus, we focus on PSAP and CSRPM.
CSRP-CDMA has the same packet format as RP-CDMA, which consists of a packet header and a packet payload.The packet header is encoded with a common code used for all nodes to transmit payload code ID placed in the header, while the packet payload is encoded with a randomly-generated payload code to transmit data portion.
The Eventual Acknowledgement policy [12] eventually acquires the receiving node to send accumulated acknowledgments (ACKs) when the receiver is idle.Because multiple transmission tasks in a cluster is very common, the opportunity of sending ACKs is not guaranteed, which can decrease the probability of packet transmission success.Hence, CSRP-CDMA uses PSAP to solve this problem above.In PSAP, the sending node divides a time frame into multiple packet segments, and waits for acknowledgements before the subsequent transmission.The receiving node when idle will send accumulated ACK packets, which are encoded with the source payload spreading code.PSAP not only guarantees the efficiency and reliability of the packet transmission, but also maintains the frame design regulations of the upper layer adapted to the existing systems.
Since the system utilizes the slotted synchronization, the length of the packet header is equal to that of one slot, and the packet header (PH) would be transmitted only at the beginning of a slot.Each node has a neighbor status table (NST) in its own register.The node only stays at send-status (SS) or receive-status (RS).When the node is transmitting a packet segment, the status is called SS; otherwise, it is called RS.
Each packet header should contain the source address (SA), destination address (DA), and the end identifier of the packet segment (PEID) or time frame (TEID).The node at RS can monitor the packet headers of neighbor nodes (NNs) and then calculate which NN will transmit packet in the next packet duration.According to these results, the node can confirm whether the destination node (DN) stays at RS if it wants to transmit packets.Then the node can decide to transmit packets immediately when the DN stays at RS, or to postpone packet duration for sensing status change of the DN again.The steps of CSRPM are as follows (Fig. 5

Performance Analysis
In this section, we calculate the expressions of packet transmission probability, normalized network throughput, and SINR.We assume that all nodes in the network are homogenous, and Power t denotes the transmitted power.The DCN consists of U fully connected clusters (  .In addition, the header spreading code is fixed until its cluster ends, while the payload spreading code used for each packet payload is generated randomly with an equal chance every time.Furthermore, all nodes are distributed uniformly and independently, and the number of the sending nodes follows the Poisson distribution.The major notations used in performance analysis are in Tab. 2.

Packet Transmission Probability
In slotted ad hoc networks, a packet accesses channels randomly, and is only transmitted at the beginning of a slot.The packet transmission will fail following the failure in the packet header transmission or the payload transmission.Hence, the success of packet transmission is defined as both packet header detection success and packet payload transmission success.Then the packet transmission probability is expressed as First, each cluster transmits its packet headers by using a unique header spreading code, which is used to distinguish the header from those of other cluster.Therefore, the reference packet header will not collide with other cluster headers.However, it still suffers from the MAIs of packet payloads transmitted near the receiver.Second, the reference packet header may collide with other headers in the same cluster since they use the same header spreading code.Thus, the probability P h of the packet header detection success is expressed as The distribution of MAI nodes near the reference receiving node Node R is shown in Fig. 6, where Node T is the reference sending node.We denote the signal propagation distance as R, and the cluster range as R c , and the radius of MAI range R max with the assumption that no interference occurs when the interference node is outside the MAI range.In the packet duration, the probability of the sending-node number k c in the cluster area S C is P s (k c ) according to (1).Furthermore, we assume that the header of the reference packet occurs in the first slot of the packet duration, so that the header will be detected successfully only when the As shown in Fig. 6, the MAI interference area S ABC of Node R can be expressed as S ABC = S A + S B + S C = R 2 max .Fig. 7 shows how the MAIs of other packets affect the reference packet headers.The Poisson probability density function (PDF) of the sending-node number k is P s (k) at rate  s according to (1).
The channel attenuation model indicates that the signal strength is attenuated badly as the distance increases.In addition, the locations of the sending nodes in the area obey a uniform and independent distribution.Therefore, we assume that the distance d i between the sending node i and Node R is within the variable set {d 1 , d 2 , …, d k }, and its PDF of which is expressed as max 2 max 2 0 ( ) 0 The derivation of (10) comes from (d 2 ) / (R 2 max ), which is the cumulative distribution function (CDF) of the distance d i .With (3), the total interference received power Powerh int at the header detector's front-end can be computed as For (11), there is no sending node in the near field with the range r, because the near field range is too small to be considered under very high frequency (VHF) tactical scenarios.Thus, we only focus on the far field.Therein, E[d i   -n ] is the expectation of signal transmission loss of all interference signals arriving at Node R .According to (10), Since Node T .must be in the reference cluster area S c , then 0 < d T  R. Similarly, the expectation ] of the transmission loss of the reference signal from Node T . is computed as According to ( 1), ( 4), ( 11), ( 12) and ( 13), the probability P hi of header detection success under interference is computed as In ( 14), K hmax can be calculated as The reference payload suffers from the MAIs of other packets in the MAI area, but it will collide with other payloads if they use the same spreading code.Node T ., which the reference payload comes from, must be within the reference cluster area S c .As previously described, the PDF of the sending-node number k in the MAI area S ABC is P s (k).From ( 3) and ( 5), Powerd int can be computed as Like P hi , the probability P d of payload transmission success with the presence of interference is expressed as In (17), K dmax can be computed as In (17), P codecol (k), the probability that the payload spreading code used for the reference packet is different from the codes used for other packets, is calculated as

Normalized Network Throughput
The network load G is defined as the average number of packets generated by the reference cluster in one packet duration.Thus, In (20), S c is the area of the reference cluster, and  s is the Poisson probability density of sending-node number in the cluster.
The network throughput S is defined as the average number of the successfully received packets in one cluster, and it satisfies S = GP p [19].Considering that the spreadspectrum networks occupy extra spectrum resources, we introduce a spread gain into S during the network performance evaluation.Hence, the normalized network throughput S normal can be defined as

SINR
The works above indicate that the expectation of sending-node number in the MAI area S ABC of Node R is calculated as Using ( 12) and ( 22), we can obtain the mean power E[I] of interference signals at the receiver's front-end as follows: Therefore, the signal interference to noise ratio SINR m is computed as According to (24), SINR is affected by the transmission distance, sending-node distribution density, and attenuation factor.

Numerical and Simulation Results
In this section, we analyze the capability of the header detector, the capability of the payload detector, S normal and SINR m .In addition, we compare CSRP-CDMA and traditional RP-CDMA on OPNET 14.5.The network scenarios considered are typical tactical VHF clustered ad hoc networks.Thus, the network parameters in this section are signal frequency f = 80 MHz, attenuation factor n = 4, signal transmission range R = 5 km, and near field range r = 300/(2f).Here, the near field is negligible.

Maximum Packet Detection Capability
We define the maximum detection capability as the maximum number of intra-cluster interference packets arriving at a receiver that will not subject the reference packet transmission to failure.This definition also implies the maximum number of concurrent packets that are recovered successfully by the detector.The maximum detection capabilities of the header detector and payload detector vs. spread-spectrum gain at different SNRs are shown in Fig. 8.The header detection capability depends on N h and the payload detection capability is limited by N d .The results show that the maximum number of received packets increases rapidly as the spreading gain grows.Moreover, the SNR of the detector's front-end affects the detection capability, but not significantly, because the AWGN is very small in comparison with MAI.Furthermore, the capability of the header detector is similar with that of the payload detector when they have the same spread gain, although the payload detection capability is slightly higher.ing header spreading gain N h and payload spreading gain N d , respectively.Based on previous works, the parameters in Fig. 9 are SNR = 10 dB, the payload to header ratio PHR = 200, as well as N d = 25 (Fig. 9a and b), and N h = 25 (Fig. 9c and d).Hence, we conclude that the payload detection capability is about 15 (Fig. 9a and b), and the header detection capability is about 15 (Fig. 9c and d).

Network Performance Analysis
All curves in Fig. 9(a) and (b) first increase and then maintain at a certain level, except the curve with 20 intra-cluster nodes.P p and S normal are optimized when the cluster contains 10 sending nodes.On the curve with 20 intra-cluster nodes, P p and S normal are close to zero when the network loads are far beyond the receiver capability.Hence, aiming at the curve with 10 intra-cluster nodes in Fig. 9 (a) and (b), we find that N h > 25 cannot affect P p or S normal , because the network performance is limited by both the header detection capability and the payload detection capability.
Similarly, on the curve with 10 intra-cluster nodes (Fig. 9c and d), P p and S normal increase when N d < 25, and S normal decreases when N d > 25.This is because the increase of N d occupies more spectrum band, which reduces S normal when N d exceeds the value required by the network load.We further derive that the network performance depends on the header and payload detection capabilities.Fig. 10 shows the network performance varying different PHR.The parameters are N h = 25, N d = 25, and SNR = 10 dB.Clearly, P p and S normal increase rapidly when PHR < 100, but stop increasing when PHR > 150 (Fig. 10a   and b).Thus, we conclude that the header should be much shorter than the payload in order to effectively reduce the collision between multiple headers.Fig. 11 shows the performance measures such as P p and S normal affected by the influence of packet size.The parameters in Fig. 11 are as follows: N h = 25, N d = 25, and SNR = 10 dB, and PHR = 200.Obviously, we can derive that the packet size does not affect the network performance.This is because each packet is transmitted in a private channel independently by using a unique payload spreading code.Moreover, P p is higher and S normal increases when the network scale is small, while they both decrease rapidly when the network scale exceeds the detection capability of the receiver.

Comparison between CSRP-CDMA and
Traditional RP-CDMA Because the PSAP and CSRPM of CSRP-CDMA are based on the traditional RP-CDMA protocol [9], Fig. 13 shows the throughputs of both CSRP-CDMA and RP-CDMA vs. network scale.The parameters in Fig. 13 are SNR = 10 dB and PHR = 200.Under the same condition, the throughput curve of CSRP-CDMA is similar with that of RP-CDMA, but CSRP-CDMA has a higher performance, because they have similar packet detection capability and different packet exchanging processes.When the network load is lower than the receiver capability, the network throughput is close to the network load, so S normal decreases as header and payload spread-spectrum gains grows.Moreover, when both gains are 25, S normal is optimal for the performance requirement of the DCN with 10 intra-cluster sending nodes.The maximum S normal of CSRP-CDMA generally outperforms the traditional RP-CDMA by 10%.

Conclusions and Futures
We proposed a CSRP-CDMA protocol to improve the control channel anti-jamming capability of a DCN.In CSRP-CDMA, the PSA policy and CSRP mechanism are presented.Considering the combination of packet header collisions, channel noise, as well as intra-cluster and inter-cluster multi-access interference, we obtained the performance analytical evaluation models for slotted spread-spectrum ad hoc networks based on CSRP-CDMA.The analytical and simulation results are listed below.1) The packet reception capability depends on both spreading gain and multiuser detector, and the network scale and packet reception capability should be well matched to optimize the network throughput.2) Compared with AWGN, the MAIs from sending nodes near the receiver affect the network performance significantly.3) The parameter configuration can be optimized from the evaluation models for DCN design.4) A compromise is needed between MAI and the requirements of cluster coverage to improve the network throughput by adjusting the transmitted power.5) CSRP-CDMA outperforms the traditional RP-CDMA by 10% when both header and payload spread-spectrum gains are 25.Hence, CSRP-CDMA is significant in the interference resistance of control channel of DCNs.Further research is needed to extend CSRP-CDMA into the DCNs with high mobility, where an important goal is to improve the interference resistance of control channel under the high mobility.
and each cluster utilizes a unique header spreading code to transmit intra-cluster headers.The header spreading codes are chosen from the code set phase.All packet payloads in the DCN are spread with the payload spreading codes generated randomly by the same code generator, and these codes are generated in the set  

Fig. 6 . 1 ck
Fig. 6.Distribution of MAI nodes nearby the reference receiving node.other

Fig. 7 .
Fig. 7. Diagram of the packet header suffer from MAI.
Gain when SNR=10dB Packet Header SS Gain when SNR=20dB Packet Header SS Gain when SNR=30dB Packet Payload SS Gain when SNR=10dB Packet Payload SS Gain when SNR=20dB Packet Payload SS Gain when SNR=30dB

Fig. 9
Fig.9shows the curves of packet transmission probability P p and normalized network throughput S normal vary-

Fig. 9 .
Fig. 9. Network performance varying header spread-spectrum gain and payload spread-spectrum gain for different network scale.(a) P p vs. N h , (b) S normal vs. N h ,.(c) P p vs. N d , (d) S normal vs. N d .

Fig. 10 .
Fig. 10.Network performances varying the ratio of Packet payload and header for different network scale.(a) P p vs. PHR.(b) S normal vs. PHR.

Fig. 12 Fig. 12 .
Fig. 12 illustrates the curves of SINR vs. network scales.The parameters in Fig. 12 are N h = 25, N d = 25, and

SNR = 10
dB, and PHR = 200.Clearly, SINR decreases when the network scale grows independent of R. It is because SINR is affected by the MAI signals arriving at the receiver rather than AWGN, and obviously, the MAI is more severe as the network scale grows.Moreover, R affects SINR largely, because the MAI signals arriving at the receiver become small as R decreases.Hence, we can adjust the transmitted power to keep R for lower MAIs.
 is the expectation of the received power of each interference packet payload.Here, we have P E ):  Step 1: Prepare to send a packet.Look up the NST for the status of DN;  Step 2: Decide the status of DN: if it is RS, go to step 9, or else, go to step 3;  Step 3: Sense and demodulate PH;  Step 4: Decide whether the PH is sent to DN: if yes, go to step 5, otherwise, go to step 8;  Step 5: Decide whether the PEID or TEID is true: if yes, go to step 6, otherwise, go to step 3;  Step 6: Wait for the next payload duration;  Step 7: Update the related node status in NST, and then go to step 9;  Step 8: Decide whether the PH comes from DN: if not, delay one slot and go to step 7, or otherwise, go to step 3;  Step 9: Send this packet.