IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 8, NO.

4, AUGUST 2019 1191

Cooperative Ambient Backscatter System: A Symbiotic

Radio Paradigm for Passive IoT
Huayan Guo , Ying-Chang Liang , Fellow, IEEE, Ruizhe Long , and Qianqian Zhang

Abstract—Cooperative ambient backscatter system is a novel

symbiotic radio paradigm, in which the passive backscatter trans-
mission shares not only the same spectrum but also the same
radio-frequency source with the active primary transmission.
By considering the unique features of the ambient backscat-
ter system, three practical symbiotic transmission schemes are
proposed, in which the relationships between the primary and
backscatter transmissions are commensal, parasitic, and com-
petitive, respectively. The achievable rates for both primary and
backscatter transmissions are derived for particular fading state.
Then, optimal power allocation is obtained for each transmission Fig. 1. The ambient backscatter communication system.
scheme under the average power constraint. Simulation results
are provided to corroborate theoretical studies.
Index Terms—Internet of Things, ambient backscatter com- the primary signal is jointly decoded and then removed at the
munications, symbiotic radio, spectrum sharing. IoT receiver (IR) by sharing some prior knowledge between
the primary and backscatter transmissions [7], [8]. In [9],
the resource allocation problem for the cooperative ambi-
ent backscatter system is investigated on the perspective of
I. I NTRODUCTION spectrum sharing philosophy.
MBIENT backscatter communication (AmBC) is a new However, in contrast to the conventional spectrum shar-
A emerging symbiotic radio technique for next-generation
Internet of Things (IoT) [1], [2]. In AmBC system, the IoT
ing model, the AmBC system has two unique features which
have not been considered in [9]. Firstly, there exists a spe-
device, also referred to as backscatter device (BD), conveys its cial signal-to-interference-plus-noise ratio (SINR) constraint
messages by riding on the radio frequency (RF) signal from at IR for the cooperative interference cancellation due to the
the primary system (e.g., cellular, TV, or WiFi systems) via RF source sharing feature. Secondly, the backscatter signal is
backscatter modulation without requiring active RF compo- much weaker than the primary signal, and the interference
nents. As a result, the backscatter transmission shares not only from the BD to primary receiver (PR) is negligible [10].
the same spectrum but also the same RF source with the pri- Therefore, it is worth reinvestigating the cooperative ambient
mary transmission, which has the potential to achieve high backscatter system by considering the above two features.
spectral and energy efficiencies [3]. In this letter, we investigate the resource allocation problem
One critical challenge for conventional AmBC is that, the for the cooperative ambient backscatter system, and three
backscatter signal from BD is very weak, while the signal practical transmission schemes are proposed with different
received directly from the primary transmitter (PT) is strong symbiotic relationships between the primary system and the
and treated as interference, resulting in quite limited transmis- backscatter system. To be specific, in commensal scheme,
sion rate [4]–[6]. Recently, the cooperative ambient backscatter the primary system may transmit independently without cater-
system has been proposed as a promising solution, in which ing the backscatter transmission. In parasitic scheme, the BD
achieves the maximum data rate, while the primary trans-
Manuscript received January 14, 2019; revised March 20, 2019; accepted mission rate is sacrificed. In competitive scheme, the two
April 11, 2019. Date of publication April 16, 2019; date of current ver- sub-systems compete for the whole system resource. We first
sion August 21, 2019. This work was supported by the National Natural
Science Foundation of China under Grant U1801261, Grant 61631005, derive the instantaneous achievable rates for particular fading
and Grant 61571100. The associate editor coordinating the review of this state. Then, closed-form solutions are derived for the transmit
paper and approving it for publication was X. Chu. (Corresponding author: power allocation problem to optimize the average achievable
Ying-Chang Liang.)
H. Guo, R. Long, and Q. Zhang are with the National Key Laboratory rates of the proposed schemes under the long-term power
of Science and Technology on Communications, University of Electronic constraint.
Science and Technology of China, Chengdu 611731, China, and also with
the Center for Intelligent Networking and Communications, University of
Electronic Science and Technology of China, Chengdu 611731, China (e-mail: II. S YSTEM M ODEL
[email protected]; [email protected]; [email protected]).
Y.-C. Liang is with the Center for Intelligent Networking and A. Channel Model
Communications, University of Electronic Science and Technology of China,
Chengdu 611731, China (e-mail: [email protected]). The cooperative ambient backscatter system is depicted
Digital Object Identifier 10.1109/LWC.2019.2911500 in Fig. 1, which consists of an active primary transmission
2162-2345 c 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on March 27,2022 at 06:04:56 UTC from IEEE Xplore. Restrictions apply
 1192 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 8, NO. 4, AUGUST 2019

(i.e., from PT to PR) and a passive backscatter transmission PT’s transmit power constraint: The average power con-
(i.e., from BD1 to IR). We consider the block fading chan- straint applies for a long-term power budget of the PT:
nel model, in which the channel coefficients remain the same E[p(ν)] ≤ Pav . (7)
for each block but may change from one block to another.
SINR constraint for SIC: The data rates of the primary
The channel coefficients of the primary channel (i.e., the PT-
transmission and the backscatter transmission are denoted by
PR channel), the direct channel (i.e., the PT-IR channel) and
Rs (ν) and Rc (ν), respectively. To achieve Rc (ν) > 0, the SIC
the backscatter channel (i.e., the PT-BD-IR channel) at fading
constraint should be guaranteed, which is represented by
state ν are represented by h0 (ν), h1 (ν) and h2 (ν), respec-
tively. Perfect channel state information on h0 (ν), h1 (ν) and γm (ν) ≤ γ1 (ν), (8)
h2 (ν) is assumed to be available to all the nodes.2 where γm (ν) is the minimum required SNR to decode s(n;
ν) and IR. The constraint in (8) is the necessary condition
B. Transmission Model for effective backscatter transmission, which however has not
Transmitted signal at the PT: Let s(n; ν) denotes the n-th been considered in [9].
symbol to be transmitted by PT at fading state ν with normal-
III. S YMBIOTIC T RANSMISSION S CHEMES AND
ized power E[|s(n; ν)|2 ] = 1, and p(ν) denotes the transmit
T RANSMISSION P OWER A LLOCATION
power. Then the transmitted signal at the PT is given by
 In this section, we propose three practical symbiotic trans-
x (n; ν) = p(ν)s(n; ν). (1) mission schemes, and then the transmission power allocation
problem is investigated under the average power constraint.
Received signal at the IR: The IR receives signal from both
the direct link and the backscatter link: A. Commensal Scheme
yI (n; ν) = h1 (ν)x (n; ν) In commensal scheme, the primary transmission achieves
 the maximum data rate, while the BD realizes effective trans-
+ ηα(ν)h2 (ν)c(n; ν)x (n; ν) + uI (n; ν), (2)
mission through adjusting the reflection coefficient α(ν) in
where c(n; ν) denotes the n-th symbol to be transmitted by BD each fading state to satisfy the SIC constraint. Both the pri-
with E[|c(n; ν)|2 ] = 1, uI (n; ν) ∼ CN (0, 1) is the additive mary signal and BD’s signal are generated from the Gaussian
noise, 0 ≤ α(ν) ≤ 1 is the normalized reflection coefficient of codebook. The transmit power of PT is optimized to maximize
BD, and η ≤ 1 denotes the backscatter efficiency. Note that, the average data rate of the primary transmission.
|h1 |2  |h2 |2 in most cases due to the double-fading effect 1) Instantaneous Achievable Rate Analysis: For particular
in the composite channel h2 [4], [6]. fading state ν, the achievable rate of primary transmission is
In this letter, we assume that IR decodes c(n; ν) via suc- Rs (ν) = log2 (1 + γ0 (ν))
cessive interference cancellation (SIC), i.e., decoding s(n; ν)
= log2 (1 + p(ν)l0 |h0 (ν)|2 ). (9)
first, and subtracting it from yI (n; ν) before decoding c(n; ν).
Therefore, the SINR for decoding s(n; ν) is given by The BD can realize effective transmission only when
the SIC constraint in (8) is satisfied. Substituting (9) and
p(ν)|h1 (ν)|2 γm (ν) = γ0 (ν) into (8), we have α(ν) ≤ ηp(ν)l
|h1 (ν)|2 −l0 |h0 (ν)|2
γ1 (ν) = . (3) 0 |h0 (ν)h2 (ν)|
2.
p(ν)α(ν)η|h2 (ν)|2 + 1 Combining 0 ≤ α(ν) ≤ 1, the optimal reflection coefficient
Then, the signal-to-noise ratio (SNR) for decoding c(n; ν) for the commensal scheme at fading state ν is
given s(n; ν) is   +
|h1 (ν)|2 − l0 |h0 (ν)|2
αc (ν) = min 1, , (10)
γ2,s (ν) = p(ν)α(ν)η|s(n; ν)h2 (ν)|2 . (4) ηp(ν)l0 |h0 (ν)h2 (ν)|2

Received signal at the PR: Since the PR is usually located where (x )+ denotes max(x, 0). Then, the achievable rate of
far away from BD, the signal backscattered from BD to PR is BD transmission is
negligible [10]. For simplicity, we ignore the signal backscat-  
Rc (ν) = Es log2 (1 + Γγ2,s (ν))
tered from BD to PR, and the PR only received signal from  
the PT: = Es log2 (1 + Γηp(ν)αc (ν)|s(n; ν)h2 (ν)|2 ) , (11)

yP (n; ν) = p(ν)l0 h0 (ν)s(n; ν) + uP (n; ν), (5) where 0 ≤ Γ ≤ 1 is the SNR-loss factor which indicates
the SNR gap to the information theoretical channel capacity,
where l0 denotes the ratio of the PT-PR channel path loss to due to the use of practical backscatter coding and modulation
the PT-IR channel path loss, and uP (n; ν) ∼ CN (0, 1) is the schemes.
additive noise. The SNR for decoding s(n; ν) is given by Since s(n; ν) is generated from the Gaussian codebook,
γ0 (ν) = p(ν)l0 |h0 (ν)|2 . (6) |s(n; ν)|2 follows standard exponential distribution, so we
have
1 The BD considered is semi-passive with an internal power source to ∞
support the low power circuit operations without using energy harvesting. Rc (ν) = log2 (1 + Γηp(ν)αc (ν)|h2 (ν)|2 x )e −x dx
2 The channel state information can be obtained by the classic channel train- 0
1 1
ing, estimation, and feedback mechanisms [5]–[7]. The impact of channel = −e β1 (ν) Ei(− ) log2 e, (12)
estimation imperfection is beyond the scope of this letter. β1 (ν)

Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on March 27,2022 at 06:04:56 UTC from IEEE Xplore. Restrictions apply
 GUO et al.: COOPERATIVE AMBIENT BACKSCATTER SYSTEM: SYMBIOTIC RADIO PARADIGM FOR PASSIVE IoT 1193

∞ 1 −z
where Ei(x )  − −x ze dz is the exponential integral, and 2) Transmit Power Allocation: The power allocation
β1 (ν) = Γηp(ν)αc (ν)|h2 (ν)|2 . problem is formulated as follows:
2) Transmit Power Allocation: The power allocation (P2) max E[Rc (ν)]
problem to maximize the average primary transmission rate p(ν)
is formulated as follows: s.t. (13a), (13b),
where Rc (ν) is shown in (15). Likewise, the optimal solution
(P1) max E[Rs (ν)]
p(ν) of (P2) is given by
 +
s.t. E[p(ν)] ≤ Pav , (13a) log2 e 1
p  (ν) = − , (18)
p(ν) ≥ 0, ∀ν, (13b) λ Γη|h2 (ν)|2
where λ can be obtained by solving E[p  (ν)] = Pav .
where Rs (ν) is shown in (9).
Theorem 1: The optimal solution of (P1) is given by C. Competitive Scheme
 + In competitive scheme, the two transmissions compete for
 log2 e 1
p (ν) = − , (14) the system resource. Gaussian codebook is employed by both
λ l0 |h0 (ν)|2 transmissions. In addition, the reflection coefficient α is fixed
where λ can be obtained by solving E[p  (ν)] = Pav . in this scheme to simplify the power allocation at PT. The SIC
Proof: The proof is straightforward and thus omitted here constraint is catered by the primary transmission, so the BD
for brevity. may increase α to improve its data rate. Then, E[Rs (ν)] is
maximized by optimizing the transmit power of PT.
1) Instantaneous Achievable Rate Analysis: To satisfy the
B. Parasitic Scheme SIC constraint, the achievable rate for primary transmission is
In parasitic scheme, the BD achieves the maximum data given by
rate, while the primary transmission rate is sacrificed. In con-
trast to the commensal scheme, the polyphase coding scheme Rs (ν) = min(Rs,1 (ν), Rs,2 (ν)), (19)
is employed by the primary transmission, while BD still where
employs the Gaussian codebook. In addition, the maximum
reflection coefficient α(ν) = 1 is adopted, and Rs (ν) is Rs,1 (ν) = log2 (1 + p(ν)l0 |h0 (ν)|2 ), (20)
adjusted to satisfy the SIC constraint in (8). Moreover, the p(ν)|h1 (ν)|2
Rs,2 (ν) = log2 (1 + ). (21)
transmit power of PT is optimized to maximize the average ηαp(ν)|h2 (ν)|2 + 1
data rate of the BD transmission.
Since s(n; ν) is generated from Gausssian codebook, the
1) Instantaneous Achievable Rate Analysis: Suppose con-
achievable rate of BD transmission is
straint (8) is satisfied, the data rate of BD transmission is  
Rc (ν) = Es [ log2 (1 + Γγ2,s (ν))]. When the primary transmis- Rc (ν) = Es log2 (1 + Γηαp(ν)|s(n; ν)h2 (ν)|2 )
sion employs the Wyner polyphase coding, i.e., |s(n; ν)|2 = 1, 1 1
Rc (ν) becomes: = −e β2 (ν) Ei(− ) log2 e, (22)
β2 (ν)
Rc (ν) = log2 (1 + Γηp(ν)|h2 (ν)|2 ). (15) where β2 (ν) = Γηαp(ν)|h2 (ν)|2 .
2) Transmit Power Allocation: The power allocation
For catering the SIC constraint, the SNR for primary
problem is formulated as follows:
transmission should satisfy
(P3) max E[Rs (ν)]
γp (ν) = min(γ0 (ν), γ1 (ν)) p(ν)
  s.t. E[p(ν)] ≤ Pav ,
p(ν)|h1 (ν)|2 (23a)
= min p(ν)l0 |h0 (ν)|2 , . (16)
p(ν)η|h2 (ν)|2 + 1 Rs,1 (ν) ≥ Rs (ν), ∀ν, (23b)
Rs,2 (ν) ≥ Rs (ν), ∀ν, (23c)
For the polyphase coding scheme,3 the achievable rate is given
by [11]: p(ν) ≥ 0, ∀ν. (23d)
∞     In (P3), only (23a) is long-term constraint, while all the
fp (τ ) 2γp (ν)
Rs (ν) = − fp (τ ) log2 dτ + log2 , other constraints are short-term. By introducing the Lagrange
0 τ e
dual variable λ (λ ≥ 0) associated with (23a), the Lagrangian
(17) of (P3) can be written as
2
where fp (τ ) = 2τ γp (ν)e −γp (ν)(1+τ ) I0 (2τ γp (ν)), and I0 (·) L(p(ν), λ) = E[Rs (ν)] − λ(E[p(ν)] − Pav ). (24)
is the zero-order modified Bessel function of the first kind.
Then the Lagrangian dual function is given by
3 In [9], the authors also assumed |s(n; ν)|2 = 1. However, the rate expres-
sion for Gaussian codebook was employed to calculate Rc (ν) for simplicity. G(λ) = max L(p(ν), λ)
p(ν)
This approximation may not be rigorous especially when the decoding SNR
is high. s.t. (23b), (23c), (23d),

Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on March 27,2022 at 06:04:56 UTC from IEEE Xplore. Restrictions apply
 1194 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 8, NO. 4, AUGUST 2019

and the dual problem is minλ≥0 G(λ). It can be verified that

the duality gap is zero, and thus solving the dual problem is
equivalent to solving the original problem (P3). Next, we focus
on obtaining G(λ) with a given λ. Then, λ can be obtained
by solving E[p  (ν)] = Pav .
Since (23b), (23c) and (23d) are all short-term constraints,
G(λ) can be obtained by solving the following problem for a
particular fading state:
(P4) max Rs − λp
p
s.t. Rs,1 ≥ Rs , Rs,2 ≥ Rs , p ≥ 0.
1 |h |2
1
Let p0 = ηα|h 2 ( l |h |2 − 1). One can check that, when p >
2| 0 0
p0 , we have Rs,1 > Rs,2 , and when p ≤ p0 , we have Rs,1 ≤
Rs,2 . Then, the above (P4) can be further decoupled into two
subproblems after some manipulations: Fig. 2. Illustration of the average achievable rate pairs of the proposed
(P4a) max Qa (p)  log2 (1 + pl0 |h0 |2 ) − λp schemes for different Pav and η.
p
s.t. p ≤ p0 , p ≥ 0,
and V. C ONCLUSION
p|h1 |2 In this letter, three practical symbiotic transmission
(P4b) max Qb (p)  log2 (1 + ) − λp schemes, i.e., commensal, parasitic and competitive schemes,
p pηα|h2 |2 + 1
s.t. p > p0 , p ≥ 0. were proposed for the cooperative ambient backscatter system.
For particular fading state, the instantaneous achievable rate
Theorem 2: The optimal solution of (P4a) is was derived for each transmission scheme. Then, closed-
pa = max(0, min(p0 , p1 )), form solutions for the transmit power allocation problem
log e
were derived subsequently under the average power constraint.
where p1 is given by p1 = ( λ2 − l |h1 |2 )+ . The optimal Finally, the average achievable rates for different schemes were
0 0
solution of (P4b) is studied and compared through numerical simulations.
pb = max(0, p0 , p2 ),
√
−(|h |2 +2ηα|h |2 )+ Δ R EFERENCES
where p2 is given by p2 = 2ηα|h1 |2 (|h |2 +ηα|h
2
2 , and
2 1 2| )
[1] V. Liu et al., “Ambient backscatter: Wireless communication out of thin
Δ = (|h1 | + 2ηα|h2 | ) − 4ηα|h2 | (|h1 | + ηα|h2 |2 )(1 −
2 2 2 2 2
air,” in Proc. ACM SIGCOMM, Aug. 2013, pp. 39–50.
|h1 |2
λ log2 e). Finally, the optimal solution of (P4) is
[2] H. Guo, Y.-C. Liang, R. Long, S. Xiao, and Q. Zhang, “Resource allo-
cation for symbiotic radio system with fading channels,” IEEE Access,
pa , if Qa (pa ) > Qb (pb ), vol. 7, pp. 34333–34347, 2019.
p = (25)
pb , otherwise. [3] L. Zhang, Y.-C. Liang, and M. Xiao, “Spectrum sharing for
Internet of Things: A survey,” IEEE Wireless Commun., early access.
Proof: The proof is omitted here for brevity. doi: 10.1109/MWC.2018.1800259.
[4] J. D. Griffin and G. D. Durgin, “Complete link budgets for backscatter-
radio and RFID systems,” IEEE Antennas Propag. Mag., vol. 51, no. 2,
IV. N UMERICAL R ESULTS pp. 11–25, Apr. 2009.
We assume i.i.d. Rayleigh fading for all channels involved. [5] J. Qian, F. Gao, G. Wang, S. Jin, and H. Zhu, “Noncoherent detec-
tions for ambient backscatter system,” IEEE Trans. Wireless Commun.,
The simulation results are obtained by averaging over 105 vol. 16, no. 3, pp. 1412–1422, Mar. 2017.
channel realizations. Fig. 2 shows the average achievable rate [6] H. Guo, Q. Zhang, S. Xiao, and Y.-C. Liang, “Exploiting multiple anten-
pairs for the three proposed schemes with different Pav and nas for cognitive ambient backscatter communication,” IEEE Internet
Things J., vol. 6, no. 1, pp. 765–775, Feb. 2019.
η where l0 = 1, Γ = 0.8, |h1 |2 = 1 and |h2 |2 = 0.01. We [7] Q. Zhang, H. Guo, Y.-C. Liang, and X. Yuan, “Constellation
set α = 1 for the competitive scheme. It is seen that, all learning-based signal detection for ambient backscatter communication
the achievable rate increases with the increase of Pav . Also, systems,” IEEE J. Sel. Areas Commun., vol. 37, no. 2, pp. 452–463,
Feb. 2019.
Rc increases when η increases. For the commensal scheme, [8] G. Yang, Q. Zhang, and Y.-C. Liang, “Cooperative ambient backscatter
Rs is irrelevant to η. However, for the competitive scheme, Rs communications for green Internet-of-Things,” IEEE Internet Things J.,
decreases when η increases, since the backscatter transmission vol. 5, no. 2, pp. 1116–1130, Apr. 2018.
[9] X. Kang, Y.-C. Liang, and J. Yang, “Riding on the primary: A
competes more system resource. Interestingly, Rs increases new spectrum sharing paradigm for wireless-powered IoT devices,”
when η increases for the parasitic scheme. Thus, the enhance- IEEE Trans. Wireless Commun., vol. 17, no. 9, pp. 6335–6347,
ment of the backscatter signal may not necessarily lead to the Sep. 2018.
[10] K. Ruttik, R. Duan, R. Jäntti, and Z. Han, “Does ambient backscat-
degradation of Rs , thanks to the proposed power allocation ter communication need additional regulations?” in Proc. DySPAN,
strategy.4 Oct. 2018, pp. 1–6.
[11] A. D. Wyner, “Bounds on communication with polyphase coding,” Bell
4 A similar phenomenon can be observed in [2, Fig. 10]. Syst. Tech. J., vol. 45, no. 4, pp. 523–559, Apr. 1966.

Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on March 27,2022 at 06:04:56 UTC from IEEE Xplore. Restrictions apply