FFT Traffic Classification-Based Dynamic Selected IP Traffic Offload Mechanism for LTE HeNB Networks

Abstract

Traffic offloading is a promising technique to alleviate the traffic load in LTE core networks. Based on 3GPP “SIPTO” (Selected IP Traffic Offload) architecture, this paper proposes dynamic SIPTO mechanism (D-SIPTO) for traffic offloading in LTE HeNB networks, which combines fast fourier transform (FFT) based IP traffic classification scheme (FFTTCS) with the dynamic traffic offload path selection algorithm (DTOPSA). Simulation results show that FFTTCS can realize on-line traffic classification with similar precisions but only using less than 10 % of the time needed by existing methods. Combined with DTOPSA, the proposed D-SIPTO can reduce the core network traffic by 60 % while selecting the optimal offload path according to the type of traffic to be offloaded.

Appendix A

According to TOPSIS [12], the algorithm to calculate $C_j^*$ is as follows.

• 1. Assume there are $m$ offload paths and $k$ QoS attributes. $x_{jl}^{(i)} = {f_{i,ql}} - {r_{j,ql}}$, $(1 \le j \le m,1 \le i \le n,1 \le l \le k)$ represents the distance between the traffic flow $i$’s QoS requirements and the offload path $j$’s QoS supply on the $l$-th QoS attribute ${q_l}$. The decision matrix is ${X^{(i)}} = {(x_{jl}^{(i)})_{m \times k}}$.

• 2. Calculate the standardized matrix ${Y^{(i)}} = {(y_{jl}^{(i)})_{m \times k}}$ with $y_{jl}^{(i)} = x_{jl}^{(i)}/\sum \nolimits _{j = 1}^m {(x_{jl}^{(i)} )^2}$ as follows:

  $$\begin{array}{lll} {Y^{(i)}} &=& {\left(y_{jl}^{(i)}\right)_{m \times k}}{ = \left(x_{jl}^{(i)}/\sum\nolimits_{j = 1}^m {\left(x_{jl}^{(i)}\right)^2}\right) }\\ &=&\left[ \begin{array}{l} y_1^{(i)}(1)y_1^{(i)}(2)...y_1^{(i)}(k)\\ y_2^{(i)}(1)y_2^{(i)}(2)...y_2^{(i)}(k)\\ ...\\ y_m^{(i)}(1)y_m^{(i)}(2)...y_m^{(i)}(k) \end{array} \right] \end{array}$$

  (5)

• 3. Determine the ideal solution ${(Y_0^{(i)})^ + }$ and negative ideal solution ${(Y_0^{(i)})^ - }$. ${(Y_0^{(i)})^ + }$ and ${(Y_0^{(i)})^ - }$ represent the ideal optimum and the ideal worst path provision ability for flow $i$.

  $$ {\left(Y_0^{(i)}\right)^ + }={\left(y_0^{(i) + }(1),y_0^{(i) + }(2),...,y_0^{(i) + }(l),...,y_0^{(i) + }(k)\right)} $$

  (6)

  where

  $$\begin{array}{@{}rcl@{}} {y_0^{(i) + }(l)}&=&\left\{\left(\mathop {\max }\limits_{1 \le j \le m} y_j^{(i)}(l)|{q_l} \in {Q^ + }\right),\right. \\ &&\left.\left(\mathop {\min }\limits_{1 \le j \le m} y_j^{(i)}(l) | {q_l} \in {Q^ - }\right)\right\}\ \end{array} $$
  $$ {\left(Y_0^{(i)}\right)^ - }={\left(y_0^{(i) - }(1),y_0^{(i) - }(2),...,y_0^{(i) - }(l),...,y_0^{(i) - }(k)\right)} $$

  (7)

  where

  $$\begin{array}{lll} {y_0^{(i) - }(l)}&=&\left\{\left(\mathop {\min }\limits_{1 \le j \le m} y_j^{(i)}(l)|{q_l} \in {Q^ + }\right),\right. \\ && \left.\left(\mathop {\max }\limits_{1 \le j \le m} y_j^{(i)}(l)|{q_l} \in {Q^ - }\right)\right\}\ \end{array} $$

  where ${Q^ + }$ is the set of QoS attributes of which a large value stands for high QoS such as SNR, while ${Q^ -}$ is the set of QoS attributes of which a small value represents good QoS such as delay.

• 4. Calculate the Euclidean distances between the $l$-th attribute of offload path $j$ to the ideal solution and to the negative ideal solution, denoted as $D_j^{(i) +}$ and $D_j^{(i) -}$, respectively, and given by

  $$ {D_j^{(i) + }}={\sqrt {\sum\limits_{l = 1}^k {{{\left[y_j^{(i)}(l) - y_0^{(i) + }(l)\right]}^2}} },(j = 1,2,...,m)} $$

  (8)
  $$ {D_j^{(i) - }}={\sqrt {\sum\limits_{l = 1}^k {{{\left[y_j^{(i)}(l) - y_0^{(i) - }(l)\right]}^2}} },(j = 1,2,...,m)} $$

  (9)

• 5. Finally, the closeness coefficient of each alternative is given by

  $$ {C_j^{(i) * }}={\frac{D_j^{(i) - }}{{D_j^{(i) - }}+{D_j^{(i) + }}} ,(j = 1,2,...,m)} $$

  (10)