To achieve efficient spectrum access for 5G networks, various influential factors must be considered in the design, such as the number of devices and the bandwidth requirement. Therefore, 5G networks have to have a much higher degree of flexibility and scalability than those of former generations. The UFMC, which is an attractive waveform for 5G networks, is vulnerable due to the issues described above. Thus, we propose an interference cancelation scheme for UFMC systems to solve this problem and present the scheme in detail.
UFMC system model
Figure 1 shows the UFMC system model and our proposed interference cancelation scheme. Compared with standard OFDM systems, the entire band of this model with N subcarriers is divided into M subbands, which correspond to M pieces of equipment. Each subband can be allocated to either one piece of equipment or physical resource block (PRB) in LTE, and each piece of equipment occupies a different amount of consecutive subcarriers determined by its service type [24]. Additionally, the subband sidelobe level can be significantly suppressed by using a bandpass filter (BPF). However, filtering has some negative effects on a certain number of subcarriers, especially on the edges of the subband. Thus, the proposed scheme is shown in Fig. 1.
The process of modulation-demodulation shown in Fig. 1, including the transmitter and receiver, is as follows. At the transmitter, the modulation control unit uses subcarrier modulation strategy to generate interference cancelation and data subcarriers to reduce the interference; then, by means of an N-point inverse discrete Fourier transform (IDFT) converter, the frequency-domain subband signal X
i
(k) is converted into a time-domain signal x
i
(n), with output length N. After the IDFT operation on each subband, the signal passes to the BPF with length L, so the length of a UFMC symbol becomes N+L−1 because of the convolution process. Both the Doppler effect due to moving equipment and local oscillator misalignment between transceivers have to be considered to model the carrier frequency offset (CFO), and the transmitted signal of the UFMC is generated by summing all filtered subband signals. From the view of the receiver, a 2N-point discrete Fourier transform (DFT) is performed after appending zeros, and a subband allocation unit is used to estimate the symbols in individual subbands. Eventually, the demodulation control unit adopts a similar strategy as that of the modulation block to complete the signal estimations for both the interference cancelation and data subcarriers. A mathematical analysis of the above process is presented in the following.
For an arbitrary ith subband B
i
(i ∈[1 : M]), the frequency domain signal X
i
(k) of the ith equipment is transformed to the time domain x
i
(n) by the IDFT, and its expression is
$$ {x}_{i}(n)=\frac{1}{N}\sum_{k\in {B}_{i}}^{}{X}_{i}(k){e}^{j\frac{2\pi}{N}nk},\quad n=0,1,\cdots,N-1 $$
(1)
Then, the complete original signal in the frequency domain X(k) is the sum of each X
i
(k)
$$ X(k)=\sum_{i=1}^{M}{X}_{i}(k) $$
(2)
By filtering through BPF, the output signal t
i
(n) is the result of discrete linear convolution between the filter impulse response f
i
(n) and the time-domain signal x
i
(n). As previously mentioned, f
i
(n) has length L, and t
i
(n) has length N+L−1. Therefore, the formula of UFMC symbol y(n), in consideration of CFO, is expressed as
$$ y(n)=\sum_{i=1}^{M}c_{i}(n)\cdot t_{i}(n)=\sum_{i=1}^{M}c_{i}(n)\cdot \left(x_{i}(n)\ast f_{i}(n)\right) $$
(3)
where c
i
(n) is the time-domain frequency-offset expression of the ith subband with the same length as t
i
(n), and * denotes the linear convolution operator. In the frequency domain, \({\hat {C}_{i}}\,({k})\) is the 2N-point DFT of c
i
(n) and can be presented as
$$\begin{array}{@{}rcl@{}} {\hat{C}}_{i}(k)&{}={}&\frac{1}{2N}\sum_{n=0}^{N+L-2}{e}^{j\frac{2\pi }{2N}\left(2\varepsilon-k\right)n} \\ &{}={}&\frac{sin\left[\frac{\pi}{2N} \left(2\varepsilon-k\right) \left(N+L-1\right)\right]}{2N\cdot sin\left[\frac{\pi }{2N}(2\varepsilon-k)\right]}\\ &&\cdot {e}\ ^{j\frac{\pi}{2N} \left(2\varepsilon-k(N+L-2)\right)} \end{array} $$
(4)
where ε denotes the relative CFO for subband i. This equation shows the frequency offset acting on subcarrier k, which is caused by CFO, that damages the orthogonality between carriers, that is, the ICI.
On the receiving end, a 2N-point DFT is used to perform the conversion from a time-domain signal to a frequency-domain signal. Then, we can derive the received symbols \(\hat {Y}\,(k)\) as
$$\begin{array}{@{}rcl@{}} \hat{Y}(k)&{}={}&\sum_{l=1}^{M}\sum_{d=0}^{2N-1}{\hat{C}}_{l}(k-d){\hat{X}}_{l}(d){\hat{F}}_{l}(d)+\hat{E}(k) \\ &{}={}&\sum_{d=0}^{2N-1}{\hat{C}}_{i}(k-d){\hat{X}}_{i}(d){\hat{F}}_{i}(d) \\ &&{+}\:\sum_{\substack{l=1 \\ l\neq i}}^{M}\sum_{d=0}^{2N-1}{\hat{C}}_{l}(k-d){\hat{X}}_{l}(d){\hat{F}}_{l}(d)+\hat{E}(k) \end{array} $$
(5)
where the signals of both \(\hat {X} _{i}\,({k})\) and \(\hat {F}_{i}\,({k})\) with period 2N are 2N-point DFTs of x
i
(n) and f
i
(n), respectively, and \(\hat {E}\,({k})\) is an additive noise sample of subcarrier k.
To gradually illustrate the relationship between N-point sequence X
i
(k) and Y
i
(k) and separate the desired signal part from the interference part in Eq. (5), we first derive \(\hat {X}_{i}\,({k})\) from Eq. (1) as follows
$$ {\hat{X}}_{i}(k)=\left\{ \begin{array}{lcc} {X}_{i}\left(\frac{k}{2}\right) & \text{if}\ k\ \text{is even} \\ \\ \sum\limits_{m\in {B}_{i}}^{}{X}_{i}(m)\frac{sin\left(\frac{\pi }{2}\left(2m-k\right)\right)}{N\,sin\left(\frac{\pi }{2N}\left(2m-k\right)\right)} \\ \qquad\quad \cdot {e}^{j\frac{\pi }{2}\left(2m-k\right)\left(1-\frac{1}{N}\right)} & \text{if}\ k\ \text{is odd} \end{array}\right. $$
(6)
Equation (6) indicates that the odd subcarriers contain part of the signal energy and the interference, which comes from other subcarriers because of the 2N-point DFT. Additionally, the 2N-point received sequence has the same conditions. By combining 2N-point signal Ŷ(k) with the relationship between N-point DFT and 2N-point DFT, we obtain the expression for N-point received signal Y(k)
$$\begin{array}{@{}rcl@{}} Y(k)&{}={}&\hat{Y}\left(\frac{m}{2}\right) \qquad \text{if}~ m=2k \\ &&m=0,1,\cdots,2N-1 \\ &&k=0,1,\cdots,N-1 \end{array} $$
(7)
Then, we consider the interference in only one subband because the ISBI from other subbands is suppressed sufficiently by filters. To simplify the analytical model, all the signals in odd subcarriers are ignored. According to Eqs. (5), (6), and (7), we separate the desired signal from the received symbols and obtain the N-point received signal of the ith subband as
$$\begin{array}{@{}rcl@{}} {Y}_{i}(k)&{}={}&\sum_{d\in {B}_{i}}^{}{C}_{i}(k-d){X}_{i}(d){F}_{i}(d)+E(k) \\ &{}={}&{C}_{i}(0){X}_{i}(k){F}_{i}(k) \\ &&{+}\:\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{C}_{i}(k-d){X}_{i}(d){F}_{i}(d)+E(k) \end{array} $$
(8)
where C
i
(k) and F
i
(k) are N-point DFTs of c
i
(n) and f
i
(n), respectively, and E(k) is the N-point representation of \(\hat {E}\)(k). In Eq. (8), the first term represents the desired signal, where C
i
(0) takes its maximum given no frequency offset. The second term indicates the interference components, where the sequence C
i
(k−d) is the ICI coefficient between the kth and dth subcarriers in the ith subband under the assumption that the kth subcarrier is the desired signal and the dth subcarrier is the interference. In other words, Eq. (8) shows that the received signal has been distorted by the existence of interference from other subcarriers.
We focus on the effects of CFO and the filter using an additive white Gaussian noise (AWGN) channel so that the sequence S
i
(k−d) is defined as the interference coefficient to explain the interference degree between the kth and dth subcarriers in the ith subband. Its influence on the system is denoted as
$$ {S}_{i}(k-d)={C}_{i}(k-d){F}_{i}(d) $$
(9)
Then, we derive the complete received symbols as
$$ Y(k)=\sum_{i=1}^{M}{Y}_{i}(k) $$
(10)
This frequency-domain signal Y(k) that has been demodulated by the receiver is treated as the X(k) of the transmitter in conventional UFMC systems.
Proposed interference cancelation scheme
Compared to OFDM, UFMC systems have greater robustness against CFO because of the introduced filters. However, our current work shows that the carriers on the two edges of the subband are influenced by the filter, which leads to degradation of system performance. Therefore, we need an interference suppression scheme to decrease the sensitivity of internal carriers to the filter.
Coding techniques have recently been used to reduce ICI. The authors in [25] proposed a reduction technique based on a geometric interpretation of the peak interference to carrier ratio (PICR) for OFDM signals and focused on the effects of CFO in OFDM systems to reduce PICR. Another coding technique, called the ICI self-cancelation scheme, was used to suppress the interference between adjacent subcarriers with simple algorithms, by modulating one data symbol onto a pair of subcarriers with predefined weighting coefficients [19, 26]. Then, the generated interference self-canceled, and the system performed much better than standard OFDM systems. Nevertheless, the redundant modulation caused a reduction in spectral efficiency of at least one half. The mentioned schemes focused on ICI; however, our target is to reduce the interference of both filters and ICI.
To avoid significant reductions in spectral efficiency, a new interference cancelation scheme is proposed by introducing an ICI cancelation scheme into UFMC systems. Based on our analysis of carriers in the affected region of the filter, we find that the greater the distance to the subband edge is, the weaker the interference of the filter. Therefore, we concentrate on the internal interference of the filter for each subband. Here, each subband is regarded as a protected object, and the interference cancelation subcarriers are inserted in pairs on the two edges. A diagram of the process in shown in Fig. 2.
In this figure, we divide each subband into three carrier blocks. The middle position is allocated to the data carriers, and the interference cancelation carriers are placed on the two edges. Each block occupies variable bandwidth to meet the flexible requirements for 5G networks because of the diversity of the access equipment (AE) and filter type. The bandwidth of each subband is reconfigurable to support diverse packet transmission efficiently. The corresponding mathematical analysis is presented in the following.
The arbitrary ith subband B
i
is divided into three parts, that is, B
i
=[Ai1,Ai2,Ai3], and the interference cancelation carriers are constrained in either Ai1 or Ai3. Simultaneously, the original signal X
i
(d) is defined to be −X
i
(d+1), e.g., X
i
(d+1)=−X
i
(d), where d∈Ai1, Ai3, and d is even. Then, the received signal, including the interference cancelation carriers in Ai1 and Ai3, becomes
$$\begin{array}{@{}rcl@{}} {}{Y'}_{i,{A}_{i1}}\left(k\right)&\,=\,&\sum_{\substack{d\in {A}_{i1} \\ d=\text{even}}}\! {X}_{i}(d)[{C}_{i}\left(k-d\right){F}_{i}\left(d\right) \\ &&{-}\:{C}_{i}\left(k-\!(d+1)\right){F}_{i}(d+1)]+{E}_{i,{A}_{i1}}\left(k\right) \end{array} $$
(11)
$$\begin{array}{@{}rcl@{}} {} {Y'}_{i,{A}_{i3}}\left(k\right)&\,=\,&\sum_{\substack{d\in {A}_{i3} \\ d= {even}}}^{}{X}_{i}(d)[{C}_{i}\left(k-d\right){F}_{i}\left(d\right) \\ &&{-}\:{C}_{i}\left(k-(d+1)\right){F}_{i}(d+1)]+{E}_{i,{A}_{i3}}(k) \end{array} $$
(12)
These two equations show that the received desired signals in these regions are disturbed by the even carriers, and the coefficient of X
i
(d) becomes an important factor in determining the strength of the interference. Thus, the previous interference coefficient in Eq. (9) becomes
$$ {} {S'}_{i}(k-d)={C}_{i}(k-d){F}_{i}(d)-{C}_{i}\left(k-(d+1)\right){F}_{i}(d+1) $$
(13)
and the remaining received signal in Ai2, which contains unmixed data carriers, is expressed as
$$ {Y'}_{i,{A}_{i2}}(k)=\sum_{d\in {A}_{i2}}^{}{X}_{i}(d){C}_{i}(k-d){F}_{i}(d)+{E}_{i,{A}_{i2}}(k) $$
(14)
Then, the whole received signal can be written as
$$ {Y'}_{i}(k)={Y'}_{i,{A}_{i1}}(k)+{Y'}_{i,{A}_{i2}}(k)+{Y'}_{i,{A}_{i3}}(k) $$
(15)
To compare with the original scheme, the desired signal of the proposed scheme is assumed to transmit on subcarrier “0” (the edge of one subband). The difference between the original |S
i
(k−d)| and the proposed |S′
i
(k−d)| is presented in Fig. 3, which is on a logarithm scale with k=0 and N=64. In Ai1 and Ai3, (a) |S′
i
(k−d)|<|S
i
(k−d)| for most of the d values and (b) the total number of interference signals is reduced to half because we include only even terms in the summation in Eqs. (11) and (12). Consequently, the interference signals in Eq. (15) are much smaller than those in Eq. (8) owing to reductions in both the number of interference signals and the amplitudes of the interference coefficients.
An interference cancelation demodulation scheme, corresponding with the modulation strategy, is used to further reduce the interference. In the modulation process, each signal on the k+1th subcarrier (k denotes an even number) is multiplied by −1 and summed with that on the kth subcarrier. Thus, in the demodulation, the desired signal in Ai1 or Ai3 is determined by the difference between Y′
i
(k) and Y′
i
(k+1), and it can be derived as
$$\begin{array}{*{20}l} {Y^{\prime\prime}}_{i,{A}_{i1}}(k) = &{Y'}_{i,{A}_{i1}}(k)-{Y'}_{i,{A}_{i1}}(k+1) \\ = &\sum_{\substack{d\in {A}_{i1} \\ d=\text{even}}}^{}{X}_{i}(d)[-{C}_{i}\left(k-(d+1)\right){F}_{i}(d+1) \\ &\ {+}\:{C}_{i}(k-d)\left({F}_{i}(d)+{F}_{i}(d+1)\right) \\ &\ {-}\:{C}_{i}\left(k-(d-1)\right){F}_{i}(d)]+{E}_{i,{A}_{i1}}(k)\\ &-{E}_{i,{A}_{i1}}(k+1) \end{array} $$
(16)
$$\begin{array}{*{20}l} {Y^{\prime\prime}}_{i,{A}_{i3}}(k) = &{Y'}_{i,{A}_{i3}}(k)-{Y'}_{i,{A}_{i3}}(k+1) \\ = &\sum_{\substack{d\in {A}_{i3} \\ d= {even}}}^{}{X}_{i}(d)[-{C}_{i}(k-(d+1)){F}_{i}(d+1) \\ &\ {+}\:{C}_{i}(k-d)\left({F}_{i}(d)+{F}_{i}(d+1)\right) \\ &\ {-}\:{C}_{i}\left(k-(d-1)\right){F}_{i}(d)]+{E}_{i,{A}_{i3}}(k) \\&-{E}_{i,{A}_{i3}}(k+1) \end{array} $$
(17)
In addition, the signal in Ai2, which does not include the interference cancelation carriers, is the same as in Eq. (14), that is,
$$ {}{Y^{\prime\prime}}_{i,{A}_{i2}}(k)=\sum_{d\in {A}_{i2}}^{}{X}_{i}(d){C}_{i}(k-d){F}_{i}(d)+{E}_{i,{A}_{i2}}(k) $$
(18)
Eventually, the estimated signal in the ith subband is denoted as
$$ {}{Y^{\prime\prime}}_{i}(k)={Y^{\prime\prime}}_{i,{A}_{i1}}(k)+{Y^{\prime\prime}}_{i,{A}_{i2}}(k)+{Y^{\prime\prime}}_{i,{A}_{i3}}(k) $$
(19)
Therefore, the whole estimated signal can be represented as
$$ Y^{\prime\prime}(k)=\sum_{i=1}^{M}{Y^{\prime\prime}}_{i}(k) $$
(20)
Following the above analysis, the corresponding interference coefficient of the estimated signal is denoted as
$$\begin{array}{@{}rcl@{}} {S^{\prime\prime}}_{i}(k-d)&{}={}&-{C}_{i}\left(k-(d+1)\right){F}_{i}(d+1) \\ &&{+}\:{C}_{i}(k-d)\left({F}_{i}(d)+{F}_{i}(d+1)\right) \\ &&{-}\:{C}_{i}\left(k-(d-1)\right){F}_{i}(d) \end{array} $$
(21)
The amplitude of |S′′
i
(k−d)| and its comparison with both |S
i
(k−d)| and |S′
i
(k−d)| are shown in Fig. 3. In this figure, we can observe that |S′
i
(k−d)| is smaller than |S
i
(k−d)| and that |S′′
i
(k−d)| is even smaller than |S′
i
(k−d)| for the majority of d. This result indicates that the proposed demodulation scheme further reduces the interference to estimate signals whose range is in Ai1 or Ai3.
The above scheme can be further validated by the carrier-to-interference power ratio (CIR) [27]. Additive noise is omitted in the process of deducing the theoretical expression for the CIR, and the sequence S(k−d) is defined to be the universal interference coefficient as
$$\begin{array}{*{20}l} S(k-d)= \left\{ \begin{array}{ccl} {S^{\prime\prime}}_{i}(k-d) & \qquad{d\!\in {A}_{i1}, {A}_{i3}} \\ {S}_{i}(k-d) & {{d\!\in {A}_{i2}}} \end{array}\right. \end{array} $$
(22)
We obtain the desired signal power on the kth subcarrier according to Eqs. (16–18) and (22),
$$\begin{array}{@{}rcl@{}} E\left[{|R(k)|}^{2}\right]&{}={}&E\left[{|{X}_{i}(k)S(0)|}^{2}\right] \\ &{}={}&E\left[{|{X}_{i}(k)|}^{2}\right]{|S(0)|}^{2} \end{array} $$
(23)
Meanwhile, the average power of the interference signal is calculated under the assumption that the transmitted data X
i
(k) have a mean of zero and are statistically independent. The average power can be represented as
$$ \begin{aligned} E\left[\!{|I(k)|}^{2}\right]\ =&\ E\left[{|\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{X}_{i}(d)S(k-d)|}^{2}\right]\\ =&\ E\left[\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{X}_{i}(d)S(k-d)\sum_{\substack{m\in {B}_{i} \\ m\neq k}}^{}{{X}_{i}}^{*}(m){S}^{*}(k-m)\right] \\ =&\ E\left[{|{X}_{i}(d)|}^{2}\right]\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{|S(k-d)|}^{2} \end{aligned} $$
(24)
Thus, the expression of CIR for subcarrier k can be derived as
$$\begin{array}{@{}rcl@{}} \text{CIR}&{}={}&\frac{E\left[{|R(k)|}^{2}\right]}{E\left[{|I(k)|}^{2}\right]} \\ &{}={}&\frac{E\left[{|{X}_{i}(k)|}^{2}\right]{|S(0)|}^{2}}{E\left[{|{X}_{i}(d)|}^{2}\right]\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{|S(k-d)|}^{2}} \end{array} $$
(25)
From Eq. (25), the CIR expression for the proposed scheme, where the desired signal is on subcarrier “0,” is derived as
$$ {CIR}=\frac{{|{S^{\prime\prime}}_{i}(0)|}^{2}}{\sum_{\substack{d\in {A}_{i1},{A}_{i3} \\ d= {even} \\ d\neq k}}^{}{|{S^{\prime\prime}}_{i}(-d)|}^{2}+\sum_{\substack{d\in {A}_{i2}}}^{}{|{S}_{i}(-d)|}^{2}} $$
(26)
and the CIR expression of the conventional UFMC system can be represented as
$$ {CIR}=\frac{{|{S}_{i}(0)|}^{2}}{\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{|{S}_{i}(-d)|}^{2}} $$
(27)
Equation (27) has the same assumption as that of Eq. (26), that is, the desired signal is on subcarrier “0”. However, to analyze the effect of the filter in the subband, we place the desired signal in the middle subband. Then, Eq. (27) becomes
$$ {CIR}=\frac{{|{S}_{i}(0)|}^{2}}{\sum_{\substack{d\in {B}_{i} \\ d\neq k}}^{}{|{S}_{i}(k-d)|}^{2}} $$
(28)
Based on Eqs. (26–28), the CIR curves of these three situations are shown in Fig. 4, which also includes the CIR of a standard OFDM system. In this figure, the conventional UFMC systems, whose desired signal is on the edge of the subband, have a greater than 4-dB CIR reduction compared with the standard OFDM systems due to the influence of the filter. If the desired signal is in the middle subband of the conventional UFMC system, its CIR is almost the same as that of standard OFDM systems. Therefore, the interference of the filter on these signal is negligible. By contrast, the proposed scheme improves more than 12 dB compared with conventional UFMC systems in the range 0<ε≤0.5, and our scheme improves 8 dB compared with the standard OFDM systems.
This analysis shows that the proposed scheme restrains the interference of the filters and improves the system performance at the receiver. Moreover, the signal-to-noise ratio of the system is enhanced because the coherent addition doubles signal level while increasing the noise level by a factor of only \(\sqrt {2}\) due to noncoherent addition.
On the other hand, the actual spectral efficiency of the proposed scheme is reduced by the utilization of the repetition coding method. Therefore, we define (a) α as the ratio of the subcarrier amount in the middle subband to that in the whole subband and (b) β as the spectral efficiency to compare with that of standard OFDM systems. It is obvious that α≤ 1. Then, β of the proposed scheme is obtained as \(\left [\alpha +(1-\alpha)\frac {1}{2}\right ]\) (b/s/Hz), and it is smaller than 1 (b/s/Hz) of the standard OFDM system. To meet the required spectral efficiency, a larger signal alphabet size can be used to increase the band utilization. For example, combining QPSK modulation with the proposed scheme increases β to [1+α] (b/s/Hz). The spectral efficiency β is also affected by the coefficient α, which is determined by the amount of edge subcarriers. This amount is related to both the type and the detailed parameters of the filter. Moreover, no complex coding methods are required for our proposed scheme, so it is easy to implement but just slightly increases the system complexity.