Design of Special Impacting Filter for Multicarrier ABPSK System

Abstract

A rather intuitive technique known as pole-zero placement is introduced to illustrate the frequency response of the special impacting filters (SIFs) with a pair of conjugate zero-poles and deduce the equation of the pole radii. Based on that, the paper proposes an iterative scheme to derive the parameters of the cascade notch filter. The cost function is determined by the cascading notch filter’s influence on impacting filters, converting the cost function’s least square problem to a filter parameters’ standard quadratic programming problem. Finally, a cascading notch SIF (CNSIF) designed to demodulate the ABPSK signals is realized.

1. Introduction

As we know, spectrum is an unrenewable resource, and the new wireless communication systems are occupying more and more transmission bandwidth. So many corporations and research institutions try to find the optimal method to utilize the spectrum. Against this background, the ultranarrow band (UNB) technology was proposed by doctor Walker. Up until now, various modulation methods have been successively proposed with the progress of UNB development, for example, variable phase shifting keying (VPSK), enhanced VPSK, very minimum shifting keying (VMSK), pulse position phase reversal keying (3PRK), missing cycle modulation (MCM), suppressed cycle modulation (SCM), and minimum sideband modulation (MSB), and so forth. The key of the demodulation is the “zero group delay” filter. The “zero group delay” means the rise time of the filter is very short and the phase changes are preserved after the “zero group delay” filter. The core unit is a quartz crystal which is too unreliable, instable, and inflexible to be mass-produced [1–3]. Wu et al. analyzed the feasibility of the UNB and proposed a high-efficiency modulation called extended BPSK (EBPSK) [4, 5]. In 2009, they proposed continuous phase EBPSK (CP-EBPSK) based on the EBPSK [6], further tightening and reducing the EPBSK power spectrum sideband. Subsequently, they imported the pseudorandom sequence and modified spectrum parameter to improve the CP-EBPSK, that is, modified CP-EBPSK (MCP-EBPSK). All of the above techniques have the same characteristic, that is “asymmetric.” They are uniformly referred to as asymmetric binary phase shift keying (ABPSK). ABPSK could increase the transport bitrate and the spectrum utilization efficiency with a more narrow band [7]. The modulation of the ABPSK signal relies on the special impacting filter (SIF), which could convert phase changes to amplitude impacts. Details of the SIF will be given in Section 3.

In recent years, multicarrier (MC) scheme has been recognized as a potential candidate for the physical layer of the wireless communication systems which could not only gain spectral efficiency but also lower out-of-band radiation. The first candidate of MC schemes is orthogonal frequency-division multiplexing (OFDM) [8]; however, a number of shortcomings of OFDM in the MC systems have been noted as well, such as large side lobes of the frequency response of the filters that characterize the subcarrier channels, high peak-to-average power ratio (PAPR), and sensitivity to carrier frequency offset. Compared with OFDM, the ABPSK allows a high level of subchannel spectral containment as Figure 1 shows. So we propose that the ABSPK technique can be used in MC schemes.

Figure 1 

Power spectrum of 1/4MCP-EBPSK signal with three subcarriers.

2. Uniform Formula of ABPSK

Denote the following.(1)Data bit duration 𝑇, it lasts 𝑁≥1 carrier wave period, that is, 𝑇=𝑁/𝑓_𝑐.(2)Phase modulation duration 𝜏, it lasts 𝐾 carrier wave period, that is, 𝜏=𝐾/𝑓_𝑐 and 𝐾≤𝑁.(3)𝑔₀ and 𝑔₁ are modulation waveforms corresponding to bit “0” and bit “1”, respectively.

Then the uniform formula of ABPSK is defined as 𝑔₀(𝑡)=𝐴sin2𝜋𝑓_𝑐𝑔𝑡,0≤𝑡<𝑇,₁(𝑡)=𝐵sin2𝜋𝑓_𝑐𝑡±𝜎,0≤𝑡<𝜏,0≤Δ≤1,0≤𝜂≤1,𝐴sin2𝜋𝑓_𝑐𝑡,𝜏≤𝑡<𝑇. (1) When the phase changes abruptly, 𝜎∈[0,𝜋]; when phase changes continuously, 𝜎=±Δsin(𝜂∗2𝜋𝑓_𝑐𝑡). The modulated waveform has the following features.(1)Data bit “0” lasts 𝑁 carrier cycles in 𝑇, the carrier cycle is 1/𝑓_𝑐 and the amplitude is 𝐴.(2)Data bit “1” consists of 𝐾 carrier cycles with an initial phase 𝜎, amplitude 𝐵 and 𝑁-𝐾 carrier cycles with the same phase as bit “0”, and amplitude 𝐴.(3)If there are no phase changes or amplitude changes, the waveforms are normal sine wave. When 𝜎∈[0,𝜋], it is the EBPSK; when 𝜎=±Δsin(𝜂∗2𝜋𝑓_𝑐𝑡), according to the sign of Δ and value of 𝜂, it is the CP-EBPSK and MCP-EBPSK, respectively.(4)When 𝐵=0, 𝜎∈[0,𝜋], (1) is MCM as has been mentioned in [9]; when 𝐵=𝐴, (1) denote the following.(a)When 𝜎=𝜋,  𝜏=𝑇, it represents the classical BPSK; when 𝜏=𝑇/4, it represents the 3PRK in [9].(b)When 𝜎=𝜋/2, 𝜏=𝑇/4, it represents 3PSK in [9].(5)The modulation degrees differ according to the phase change 𝑆 and modulate duration. The phase changes could control the bandwidth: when the phase vanishingly changes, the spectrum is the narrowest, and when the changes are 𝜋, the spectrum is the widest; Denote 𝜏/𝑇=𝐾/𝑁 as “modulated duty cycle.”

According to (1), choosing a proper value of Δ, 𝜂, 𝜎, 𝜏 and using a better shaping-filter could tighten the bandwidth of the modulated signal. We call the signal defined by (1) the uniformly asymmetric binary phase shift keying (ABPSK).

3. Special Impacting Filter (SIF)

The demodulation of AEBPSK signal based on the special digital impacting filter is in [10, 11]. The SIF is a kind of digital IIR filters; when working on the proper frequency range, it can convert the phase changes to amplitude impacting. The places where phase is unchanged have much lower impacting than the changed places and most noise is removed at the same time (see Figure 2). Based on the output response, direct amplitude detection can be used for the symbol judgment in intermediate frequency instead of down-conversion to baseband. The filtering mechanism in that filter offers different band to signal and noise, which is discussed in [10, 11] in detail.

Figure 2 

Modulated and demodulated waveform of EBPSK.

Different from normal multicarrier transmission system, the ABPSK subcarriers need not guard interval or satisfy the orthogonal condition. Every subchannel set differs from the others. A narrow pass band of the SIF is used to separate the subcarriers. In order to eliminate the adjacent channel interference, we need to set the notch frequency of the cascade notch filter the same as other carriers frequency to suppress other subchannels’ interference and the output demodulation results [10]. The tiny phase change of the ABPSK makes the demodulation performance closely related to the sampling rate. Only by setting the proper parameter of the filter could it reflect the signal change. Based on the above condition, the parameters of the SIF bank correspond with each subcarrier’s sampling rate, respectively.

4. Design of Cascade Notch SIF (CNSIF)

The key to demodulate the multicarrier ABPSK signal is the design of the receive filter bank, that is, cascade notch filter based on the SIF. Each subchannel regards the other subchannels as a sinusoidal interference, the notch filter cancelling the interference and SIF demodulating the corresponding signal. Many methods have been mentioned in [10, 11] for the SIF design and performance analysis, but the SIF pole radius is specified by the designer arbitrarily. Here we present a new approach to the numerical solution of the SIF as well as use an iterative scheme to gain the pole angle of the notch filter that fulfills the cost function.

4.1. Calculation of the SIF Pole Radii

Consider an SIF with one pair of conjugate zero-poles whose transfer function is 𝐻(𝑧)=1−𝑟_𝑧𝑒^{𝑗𝜔𝑐−𝜀}𝑧⁻¹1−𝑟_𝑧𝑒^{−𝑗𝜔𝑐−𝜀}𝑧⁻¹1−𝑟_𝑝𝑒^{𝑗𝜔𝑐+𝜇}𝑧⁻¹1−𝑟_𝑝𝑒^{𝑗𝜔𝑐+𝜇}𝑧⁻¹. (2)

Different from the conventional IIR filters, the zero angles are less than the corresponding carrier frequency and are less than the pole angles. Place the zeros on the unit circle with angle shift 𝜈, that is, 𝑟_𝑧=1, 𝜔_𝑧=𝜔_𝑐−𝜈; the poles are close to the unit circle and are near to zeros with angle shift 𝜇, that is, 0≪𝑟_𝑝<1, 𝜔_𝑝=𝜔_𝑐+𝜇. Generally the carrier frequency is situated in the maximum slope of the magnitude response. Figure 3 shows the amplitude-frequency response and phase-frequency response of the demodulation impacting filter with one pair of conjugate zero-poles, the normalized carrier frequency is 0.2𝜋 rad/sample; it has narrow notch-frequency-selecting performance near the carrier frequency.

Figure 3 

Frequency response of the SIF.

According to (2), we can evaluate the frequency response with graphical method (see Figure 4). In practice the poles and zeros are even much closer than Figure 4; the poles are even much closer to the unit circle. In the z-plane, the zero-pole vectors shift on the unit circle 𝑒^𝑗𝜔.

Figure 4 

Schematic of zero-pole relative position.

𝑧₁, 𝑧₂ are zeros lying on the unit circle; 𝑝₁, 𝑝₂ are poles near the unit circle and zeros. 𝐴₁, 𝜓₁, 𝐴₂, 𝜓₂, 𝐵₁, 𝜃₁, 𝐵₂, 𝜃₂ represent the module and angle of 𝑧₁, 𝑧₂, 𝑝₁, 𝑝₂, respectively. For the periodic frequency characteristic of the digital filters, we could just discuss the situation of 0<𝜔<𝜔_𝑠/2, that is, 0∼𝜋.

From Figure 4, the frequency magnitude response of (2) can be rewritten as 𝐻𝑒^𝑗𝜔=𝐴₁⋅𝐴₂𝐵₁⋅𝐵₂. (3)

Because the zeros and poles are close to the unit circle and the distances from the conjugate pole and zero to 𝑒^𝑗𝜔 are nearly equal when they shift in the region of 0<𝜔<𝜔_𝑧 and 𝜔_𝑝<𝜔<𝜔_𝑠/2; that is, 𝐴₁≈𝐵₁, 𝐴₂≈𝐵₂. So 𝐻𝑒^𝑗𝜔=1,0<𝜔<𝜔_𝑧∪𝜔_𝑝𝜔<𝜔<^𝑠2. (4)

But when they shift in the region of 𝜔_𝑧<𝜔<𝜔_𝑝, the distance 𝐴₁, 𝐵₁ is much smaller than the distance 𝐴₂, 𝐵₂; we may state that 𝐴₂≈𝐵₂. The frequency magnitude response can be written as 𝐻𝑒^𝑗𝜔=𝐴₁𝐵₁,𝜔_𝑧<𝜔<𝜔_𝑝. (5)

Enlarge the region of 𝜔_𝑧<𝜔<𝜔_𝑝, the relative position between zero-pole vector and 𝑒^𝑗𝜔 depicted in Figure 5, here the influence of the complex conjugate zero-pole pair will be negligible.

Figure 5 

Schematic of the enlarged relative position between zero-pole.

With the Pythagorean theorem, we obtain two equations: 𝐴₁=2sin𝜔−𝜔_𝑧2𝐵₁,=ℎ²+𝑏²=𝑟²_𝑝−𝑎²+𝑏², (6)(7) where 𝑟_𝑝 is the module of pole 𝑝₁.

Substituting 𝑎=𝑟²_𝑝−𝑟²_𝑝sin²(𝜔_𝑝−𝜔), 𝑎+𝑏=1, into (7) leads to 𝐵₁=𝑟²_𝑝−2𝑟_𝑝cos𝜔_𝑝−𝜔+1. (8)

So the frequency magnitude response in the region of 𝜔_𝑧<𝜔<𝜔_𝑝 can be written as 𝐻𝑒^𝑗𝜔=2sin𝜔−𝜔_𝑧/2𝑟²_𝑝−2𝑟_𝑝cos𝜔_𝑝−𝜔+1,𝜔_𝑧<𝜔<𝜔_𝑝. (9)

According to Moorer’s bandwidth definition [12], if the resonance gain over the DC gain exceeds 6 dB in absolute value, the frequencies at 3 dB below/above the peak/notch is the band edges, that is 𝐻𝑒^{𝑗(𝜔𝑐−𝐵𝑤/2)}=1√2⋅𝐻𝑒^𝑗𝜔𝑐, (10) where 𝐵_𝑤 is the bandwidth of the SIF; (10) can be expressed as sin𝜔_𝑝−𝜔_𝑧/2sin𝜔_𝑝−𝐵_𝑤/2−𝜔_𝑧⋅/2𝑟_𝑝²−2𝑟_𝑝cos𝐵_𝑤/2+11−𝑟_𝑝=√2. (11)

To ensure the impacting effect, the pole and zero angle is specified by the designer. When the bandwidth 𝐵_𝑤 is known, the results of the following two equations are constants, respectively, that is 𝜔_𝑝−𝜔_𝑧2=𝜙,1−cos𝜙−𝐵_𝑤/21−cos𝜙=𝑀. (12)

Substituting (12) into (11), we can readily obtain 𝑟_𝑝=𝑀−cos𝐵_𝑤/2±cos²𝐵_𝑤/2−2𝑀cos𝐵_𝑤/2+2𝑀−1.𝑀−1 (13)

The poles and zeros are very close to each other, as a consequence, 𝜙 and 𝐵_𝑤 will be relatively small. To avoid 𝑟_𝑝>1, we abandon the adding situation. After some further simplification. The final result is 𝑟_𝑝=𝑀−cos𝐵_𝑤/2−cos²𝐵_𝑤/2−2𝑀cos𝐵_𝑤/2+2𝑀−1.𝑀−1 (14)

Equation (14) shows that the pole radii of SIF can be given in a quantitative way instead of arbitrary setting.

The procedure of the one-pair conjugate zero-poles SIF design could ensure the impacting effect, and we know that if the zeros and poles are far away from the unit circle, they will only affect the amplitude of the frequency response, but not the shape. From this standpoint, we can add more conjugate poles to tighten the bandwidth of the SIF and gain higher impacting effect. Generally we can add two conjugate poles whose angles are equal to the initial poles and are far away from the unit circle.

4.2. The Cascade Notch SIF (CNSIF)

The multicarrier ABPSK scheme uses the CNSIF banks to separate and demodulate each subcarrier [13]. The key of the CNSIF is the narrow notch-frequency-selecting performance. When the overlapping signals pass the filter banks, the primary signal is enhanced and the noises or interferences are attenuated. Here the SIF enhances the primary signal and the notch filter decreases the disturbing signal. Let us start with the design of the notch filter in details.

Without loss of generality, we assume that the notch filter has 𝑘 notch frequencies; the transfer function is given by 𝐻₁(𝑧)=𝐵(𝑧)𝐴(𝑧)=∏^𝑘_𝑖=11−2cos𝜔_𝑁𝑖𝑧⁻¹+𝑧⁻²∏^𝑘_𝑖=11−2𝑟_𝑖cos𝜔_𝑝𝑖𝑧⁻¹+𝑟_𝑖²𝑧⁻², (15) where 𝑟_𝑖 is the pole radius. For stability and according to system requirements, the values of 𝜔_𝑁𝑖 and 𝑟_𝑖 are given by the designer (0≤𝑟_𝑖<1). Meanwhile the zeros are constrained to locate on the unit circle at the notch frequencies 𝜔_𝑁𝑖, so the problem of designing a notch filter reduces to find the optimal value of 𝜔_𝑝𝑖.

Let 𝑎_𝑖=2cos(𝜔_𝑝𝑖), then 𝜔_𝑝𝑖=arccos(𝑎_𝑖/2), −2≤𝑎_𝑖≤2, −𝜋≤𝜔_𝑝𝑖≤𝜋. The transfer function can be rewritten as 𝐻₁(𝑧)=∏^𝑘_𝑖=11−𝑏_𝑖𝑧⁻¹+𝑧⁻²∏^𝑘_𝑖=11−𝑟_𝑖𝑎_𝑖𝑧⁻¹+𝑟_𝑖²𝑧⁻², (16) where 𝑏_𝑖=2cos(𝜔_𝑁𝑖), 𝑖=1,2,…,𝑘.

For convenience, let 𝑖=1, then the notch frequency is 𝜔_𝑁1, pole radius is 𝑟₁, and (16) can be expressed as 𝐻₁(𝑧)=𝐵(𝑧)𝐴(𝑧)=1−𝑏₁⋅𝑧⁻¹+𝑧⁻²1−𝑎₁𝑟₁𝑧⁻¹+𝑟₁²𝑧⁻². (17)

Assuming that 𝜔_𝑧<𝜔_𝑝<𝜔_𝑁1<𝜔_𝑁2<⋯<𝜔_𝑁𝑖, 𝜔_𝑁1≥𝜔_𝑝+𝐵_𝑤/2. With a given weighting function 𝑊(𝜔), we can define the cost function as 𝐶(𝑎)=_𝑅𝑊(𝜔)𝐻𝑒^𝑗𝜔−𝐻₁𝑒^𝑗𝜔2𝑑𝜔, (18) where the region 𝑅=[𝜔_𝑝,𝜔_𝑁1−𝛿]∪[𝜔_𝑁1+𝛿,𝜋], 𝛿 is a prescribed small positive number, combining (4) and (18) yields 𝐶(𝑎)=_𝑅𝑊(𝜔)1−𝐻₁𝑒^𝑗𝜔2𝑑𝜔. (19)

To ensure the impacting characteristic of SIF and cancel the signal disturbing, the cascade notch filter should not affect the frequency response of the SIF; that is, the cost function should be minimized.

Substituting (17) into (19), we can obtain 𝐶𝑎₁=_𝑅𝑊(𝜔𝐵𝑒)1−^𝑗𝜔𝐴𝑒^𝑗𝜔2𝑑𝜔=_𝑅𝑊(𝜔)𝐴𝑒^𝑗𝜔2𝐴𝑒^𝑗𝜔−𝐵𝑒^𝑗𝜔2𝑑𝜔=_𝑅𝑊(𝜔)𝐴𝑒^𝑗𝜔2𝑟²₁−1𝑒^−2𝑗𝜔+𝑏₁𝑒^−𝑗𝜔−𝑟₁𝑎₁𝑒^−𝑗𝜔2𝑑𝜔. (20)

Let 𝑊(𝜔)/|𝐴(𝑒^𝑗𝜔)|² be the new weighting function, since 𝑟₁ and 𝑏₁ are known numbers, we need to find 𝑎₁ to minimize the cost function. For 𝑎₁=2cos(𝜔_𝑝1), 𝜔_𝑁1−𝛿≤𝜔_𝑝1≤𝜔_𝑁1+𝛿, we can use an iteration scheme to find the optimal value of 𝑎₁; each iteration refreshes the value of 𝐴(𝑒^𝑗𝜔). We denote the value of 𝑎₁ at the nth iteration as 𝑎_1𝑛, 𝐴(𝑒^𝑗𝜔) as 𝐴_𝑛(𝑒^𝑗𝜔), similarly. When 𝐴(𝑒^𝑗𝜔) gets the new value, the weighting function is updated immediately.

For convenience, substitute 𝑝(𝜔)⟵𝑟₁²−1𝑒^−2𝑗𝜔+𝑏₁𝑒^−𝑗𝜔,𝑞(𝜔)⟵−𝑟₁𝑒^−𝑗𝜔⟹𝐶𝑎_1𝑛=_𝑅𝑊(𝜔)𝐴_𝑛𝑒^𝑗𝜔2𝑝(𝜔)+𝑞(𝜔)𝑎_1𝑛²=𝑑𝜔_𝑅𝑊(𝜔)𝐴_𝑛𝑒^𝑗𝜔2⋅𝑎²_1𝑛𝑞(𝜔)²+2𝑎_1𝑛𝑝(𝜔)𝑞(𝜔+𝑝(𝜔))²=𝑑𝜔_𝑅𝑊(𝜔)𝐴_𝑛𝑒^𝑗𝜔2𝑎²_1𝑛𝑞(𝜔)²+2𝑎_1𝑛⋅Re𝑝(𝜔)𝑞^∗(𝜔)+𝑝(𝜔)²𝑑𝜔. (21)

After some simplification, (21) is rewritten as 𝐶𝑎_1𝑛=_𝑅𝑊_𝑛(𝜔)𝑞(𝜔)²⋅𝑎²_1𝑛+2⋅Re𝑝(𝜔)𝑞^∗(𝜔)⋅𝑎_1𝑛+𝑝(𝜔)²𝑑𝜔, (22) where 𝑊_𝑛(𝜔)=𝑊(𝜔)𝐴_𝑛−1𝑒^𝑗𝜔. (23)

So we can determine the value of 𝑎_1𝑛 by solving the constrained optimization problem min_𝑅𝑊_𝑛(𝜔)𝑑𝜔⋅𝑞(𝜔)²⋅𝑎²_1𝑛+2⋅Re𝑝(𝜔)𝑞^∗(𝜔)⋅𝑎_1𝑛+𝑝(𝜔)²s.t.0≤𝑟₁<1−2≤𝑎_1𝑛𝜔≤2_𝑁1−𝛿≤𝜔_𝑝1≤𝜔_𝑁1+𝛿. (24)

We can see that the cost function is a parabola with its mouth opened up, so it could be transformed to a closed form and have an optimal solution. At the nth iteration, minimizing (22) subjected to constraints of (24) is a typical quadratic programming problem. We can compute the solution efficiently.

5. Design Example

5.1. Design Parameters

Set the normalized three notch frequencies as 0.21𝜋, 0.25𝜋, and 0.3𝜋. Easily derive the initial values that 𝑎₁=0.521292, 𝑎₂=0.555360, 𝑎₃=0.587785. Let 𝛿=0.001𝜋; the convergent values are 0.52206411, 0.55613211, and 0.58855711. Given the following specifications: 𝑟₁=𝑟₂=𝑟₃=0.95; weighting function 𝑊(𝜔)=1; 𝜔_𝑧=0.6282 rad, 𝜔_𝑝=0.62855 rad, 𝐵_𝑤=𝜔_𝑝−𝜔_𝑧=0.00035 rad (1.114e⁻⁴𝜋 rad). The pole radii 𝑟_𝑝 defined in (14) is 0.999999969375. Adding another three pairs of conjugate poles with angle 0.62855rad, radii are 0.9, 0.92, and 0.95, respectively. The frequency magnitude response of the final designed CNSIF is shown in Figure 6.

Figure 6 

The frequency magnitude response of the designed CNSIF.

5.2. Performance

We choose the MCP-EBPSK signal as a source, simulating the performance of two subcarriers. The parameters are as follows: 𝑓_𝑐1=10MHz, 𝑓_𝑐2=10.04MHz, 𝑓_𝑠1=160MHz, 𝑓_𝑠2=100.4MHz, 𝐴=1; 𝐵=1; Δ=0.1; 𝜂=0.5; 𝐾=2; 𝑁₁=100; 𝑁₂=160; the entire test symbols are 3 million. Set the notch frequencies as 0.1255𝜋 and 0.1992𝜋, we can obtain 𝑎₁=1.84656, 𝑎₂=1.62098, after six iterations converge to 1.84855455 and 1.6219734, respectively. Let 𝑟₁=𝑟₂=0.95, 𝛿=0.001𝜋, weighting function 𝑊(𝜔)=1.

We can obtain the parameters of the CNSIF bank as𝑎₁= [1, −6.096111877753762, 16.418167515832877, −25.326641789930786, 24.298560015038895, −14.721898165092053, 5.447828484314318, −1.116449367584231, 0.098333159057181];𝑏₁= [0,0,0,0,1, −3.694313616799019, 5.411987912102056, −3.694313616799019,1];𝑎₂= [1, −5.406525327518239, 13.652579283244823, −20.517799618098030, 19.861352592576445, −12.536073668382514, 4.996805854728122, −1.139889481405723, 0.120047843082279];𝑏₂= [0,0,0,0,1, −3.239006288658281, 4.622788277906668, −3.239006288658281,1].

Figure 7 shows the waveform of the component signal after the CNSIF bank filtering. We can see that the designed CNSIF could separate the component MCP-EBPSK signals. The BER performance is shown in Figure 8.

Figure 7 

The waveform after the CNSIF bank filtering.

Figure 8 

BER performances of two subcarriers.

6. Conclusions

With the intuitive method known as zero-pole placement, we analyzed the frequency response of the SIF in a quantitative way. The equation about the pole radii under specified bandwidth is derived. The designed CNSIF could not only maintain the impacting characteristic but also attenuate the interchannel interference (ICI). From Figure 6, we can also see that adding notch zeros could speed up the decaying of the response curve and tighten the bandwidth; this will help to filter the noise. Compared with OFDM, the multicarrier scheme of ABPSK could have quite low PAPR ratio. It can also be combined with MIMO, cognitive radio, and so forth to get better performance, which is our next work to emphasize.

Acknowledgments

The authors thank all of the reviewers for their valuable comments, which have considerably helped in improving the overall quality of the work presented in the revised paper. Editorial assistance and proofreading of the manuscript from Bingcheng Zhu are appreciated. This work is supported by the State 863 Project (2008AA01Z227), the National Natural Science Foundation of China (NSFC), under the Grant 61271204.