- Photonics Research
- Vol. 11, Issue 8, 1449 (2023)
Abstract
1. INTRODUCTION
A quantum network is an interconnected network that makes use of the properties of quanta to transmit data, and could revolutionize ways of information exchange in the future [1]. The quantum network [2] has many breakthrough applications, such as a quantum secure network through quantum key distribution (QKD) [3–10] or quantum secure direct communication [11,12]. In addition, it can also perform tasks impossible in classical physics, such as distributed quantum computing [13] and accurate global timing [14], whcih will bring accurate navigation and Earth sensing. Moreover, the quantum network could even lead to accurate telescopes [15] and new fundamental tests of quantum nonlocality, quantum teleportation [16], and quantum gravity [17].
QKD is one prominent application of quantum networks, and is the core technology of quantum secure communication [18]. It can provide secure keys for legal parties even in the presence of eavesdroppers, and its theoretical security is guaranteed by the basic principles of quantum mechanics [19]. The QKD protocol can be divided into discrete-variable QKD (DV-QKD) [20] and continuous-variable QKD (CV-QKD) [21–42] according to the physical quantity that carries key information. According to the quantum technology roadmap released by OIDA [43], point-to-point QKD has been matured to build networks and gradually commercialized [44–64]. A representative quantum secure network is the Beijing–Shanghai trunk line [65], which achieves quite long distance transmission. In addition, the Cambridge quantum metropolitan area network is constructed with high bandwidth data transmission [66]. Furthermore, quantum networks in the United Kingdom have been operating for several years with three nodes separated by 5–10 km optical fiber [67]. The 46-node quantum metropolitan area network in Hefei realizes real-time voice telephone, text messaging, and file transmission [68]. For multi-user access, the quantum access network (QAN) proposed by Fröhlich
The physical implementation of a quantum network for QKD is an important issue. On one hand, constructors need to consider the coverage of the quantum network, which can be divided into backbone networks, metropolitan area networks, and access networks. On the other hand, builders should also be concerned with quantum network topologies, such as star, tree, and mesh. For different coverages and topologies, the physical structure of the quantum network is a vital issue. Considering the preparation and measurement of quantum states by multi-users, the quantum states generated from different sources require a large number of detectors, while those generated from the same source will interfere with each other and be difficult to separate. An alternative scheme is to build large-scale networks based on the mature point-to-point QKD system. However, limited by its complexity, this scheme cannot support access services for such a large number of users. Therefore, a new type of quantum network architecture needs to be proposed to simplify the original complex network with multi-user access.
Sign up for Photonics Research TOC. Get the latest issue of Photonics Research delivered right to you!Sign up now
Therefore, we propose an expandable and cost-effective quantum network physical structure, called round-trip multi-band QAN (RM-QAN), to improve network performance and support multi-user access. The RM network structure utilizes quantum states traveling in a circle to transmit data, and uses the frequency division multiplexing (FDM) technique to isolate different users. In RM-QAN, each user requires only one modulator and one circulator when plugging into the network. In addition, only one laser and one detector are needed to build the entire network, which is flexible and cost effective. Such networks could theoretically support multi-user access without performance penalty, and network users can achieve 600 bit/s level quantum secure key generation under 30 km network range. Moreover, network scalability and noise suppression are excellent, which is a promising solution for building near-term multi-user quantum secure networks.
In this paper, our RM-QAN scheme is introduced in detail. First, we describe the physical structure of RM-QAN and expound upon its advantages. In addition, we use RM-QAN to realize CV-QKD and evaluate the performance of QKD through a complete noise analysis. Based on this physical structure, we construct a proof-of-principle experimental platform and verify the feasibility of multi-user secure key sharing. Finally, we provide a conclusion.
2. RESULT
A. Physical Structure of RM-QAN
Our RM-QAN physical structure is described as follows. In the QAN, the quantum network unit (QNU) is the device held by the user, which corresponds to the optical network unit (ONU) [70–72] in the optical access network. The QNU does not need to be responsible for receiving light in our scheme, nor does it need to be responsible for generating light sources. Instead, QNU needs only to modulate the key information. It reduces the overall system cost of the QAN. The work of generating and receiving light is all done by the quantum line terminal (QLT). QLT is the terminal equipment used to connect quantum trunks, which corresponds to the optical line terminal (OLT) [70–72] in the optical access network. Notably, QLT serves as a terminal in the QAN, distinct from its function as a relay. In this scheme, multiple QNUs share their respective keys with the same QLT. The round-trip structure is divided into two stages. First, the optical carrier is transmitted from the QLT side to the QNU side. There is a loss of 1/N in this part, but it is not involved in the transmittance of QKD. Second, the optical carrier is transmitted from the QNU side to the QLT side. Light transmitted to the QLT side is brought together from signals of all users. There is also a loss of 1/N in this part, which is involved in the transmittance of QKD. Therefore, the multi-user quantum network can be completed with one laser and one detector, which is more efficient than other schemes. In our scheme, according to the upstream transmission direction, it can be understood that QNU is Alice and QLT is Bob in the classical scheme.
For a single user, this QKD scheme is actually a plug-and-play QKD architecture. The plug-and-play scheme was proposed in Ref. [73] and experimentally verified in DV-QKD [74–76] and CV-QKD [77]. In addition, the plug-and-play scheme has corresponding commercial QKD systems [78]. For security, QNU can further randomize the global phase of each pulse, measure the incoming intensity, and introduce sufficient attenuation, and the standard security proofs of a one-way system can be used [79], which has been used in the CV case [77].
FDM is a multiplexing technology that modulates multiple baseband signals to different frequency carriers and then superimposes them to form a composite signal. The FDM method is like a “frequency modulation (FM) radio system,” which can be tuned to different frequencies to receive information from different sources. In the implementation, since the quadrature components can be modulated on different frequency bands, they can be used to transmit the quantum key [52,53], and therefore the FDM method can be adopted for multi-user key distribution. Time division multiplexing (TDM) is an alternative scheme, but it has a high requirement of time slot control. The principle of our scheme is described in detail below.
As can be seen from the schematic diagram of RM-QAN in Fig. 1, there are multiple QNUs corresponding to one QLT. First, a continuous wave is generated and transmitted by QLT in the optical layer. Then, it is divided into pieces through an N:1 splitter, which does not carry any data. After receiving light, QNU will conduct secure key modulation through a radio frequency (RF) signal, which carries the information on different frequency bands, and can be distinguished clearly on the spectrum. QNU selects the number to form a random sequence of length with equal probability. Then QNU prepares coherent states according to the random sequence. The th coherent state can be expressed as
Figure 1.Schematic diagram of the round-trip multi-band quantum access network (RM-QAN). First, a continuous wave is generated and transmitted by the quantum line terminal (QLT). Then, the continuous wave is divided into
After receiving through a coherent receiver, QLT will get a mixed multi-band spectrum, in which the information is indistinguishable in the time domain but can be clearly distinguished in the frequency domain. QLT first checks all bands that QNU has registered. When accessing the network, each QNU needs to register its frequency band, i.e., carrier frequency. QLT checks the registered bands to see which QNU is currently communicating. This registration method can effectively prevent Eve from using illegal frequency bands to obtain information. For these bands, QLT uses bandpass filtering to separate them. Then, QLT performs the first phase shift recovery for the information, which addresses the optical phase drift during signal transmission. After that, since the key obtained at this time still carries out spectrum shifting, QLT needs to remove its carrier by coherent demodulation for each QNU. Coherent demodulation can obtain baseband signals and . The formula of coherent demodulation is
To prove that the interference between each user’s signal does not exist, assume that
This round-trip structure is mainly vulnerable to Eve’s practical security attacks, including the phase remapping attack [80,81] and the Trojan-horse attack [79]. For the phase remapping attack, the effective solutions are that QNU checks the arrival time of the reference pulse and the signal pulse by monitoring, verifying that she is applying the correct modulations to her states [80]. For the Trojan-horse attack, since we cannot use the isolator in the two-way structure, the filter can be used to exclude the eavesdropper’s input light [79]. In addition, there are three technical countermeasures: (1) installing a watchdog detector with a switch at the entrance of QLT that randomly routes a small fraction of incoming signals to this detector; (2) opening the door for Eve for a smaller time duration; (3) reducing the width of phase modulation voltage pulse [82]. Certainly, from a theoretical perspective, a higher amount of privacy amplification can help QNU and QLT to destroy the partial information of Eve. We need to estimate the maximum leakage due to Trojan-horse attacks, and incorporate these elements in the security proof [79,83,84].
The RM physical structure has many advantages compared with downstream and upstream QANs. In the downstream QAN, the transmitters are located at QLT, while the receivers are located at QNU [85]. This downstream scheme has two major disadvantages. First, each QNU in the network requires a detector, which is normally expensive and difficult to operate. Second, it is impossible to locate the data of each user certainly. Therefore, all detectors must run at the same speed as the transmitter to avoid missing the key, which means that most of the detector’s bandwidth is unused. In the upstream QAN, QNU is responsible for transmitting the key, and QLT is responsible for receiving the key [69]. The upstream scheme still requires multiple laser sources to prepare a quantum signal, which is difficult for ordinary users to afford. In addition, the network capacity has to be determined before building the upstream QAN.
RM-QAN solves the above issues well. In our scheme, each QNU needs only one modulator and one circulator to plug into the network, and the entire network requires only one laser and one detector, which is efficient. Moreover, the bandwidth of the detector is fully used due to FDM. In addition, network scalability and noise suppression are wonderful, which means a large number of users can easily access it at any time.
B. CV-QKD Based on RM-QAN
1. Secret Key Rate
In the following, we evaluate the reachable secret key rate under the RM-QAN for discrete modulation coherent state (DMCS) CV-QKD [34]. The formula of the secret key rate for unit system repetition frequency is in Appendix A. For the practical CV-QKD system, the secret key rate can be calculated as
In Eq. (7), represents the noise introduced by Rayleigh backscattering. denotes the noise introduced by the frequency cross talk, which is caused by photons leaking from other frequency bands. In addition, represents the noise introduced by the imperfection of the optical circulator, especially concerning the isolation and directionality. is the modulation noise caused by the uncertainty of modulation voltage. The change in the number of photons caused by spontaneous radiation will be reflected in the amplitude of the laser, forming amplitude noise . The spontaneous radiation of the laser causes not only a change of intensity but also a random change in the frequency of the laser pulse signal, forming phase noise . Specific theoretical derivation values of each noise component can be found in Section 3.A.
Other parameters are quantum efficiency , electrical noise , reconciliation efficiency , modulation variance , and repetition frequency . The N:1 splitter on the return path can introduce 1/N loss on each arm, thereby reducing the secret key rate of all users, and transmittance will change to , where is the branch number of the N:1 splitter. In the next step, we can introduce high speed optical switches to eliminate the loss caused by the splitter. The comparison diagram of the secret key rate between this scheme of different network capacities and other classical schemes is shown as Fig. 2.
Figure 2.Comparison diagram of secret key rate between this scheme of different network capacities and other classical schemes. Parameters are set as
In Fig. 2, the Pirandola–Laurenza–Ottaviani–Banchi (PLOB) bound has the farthest transmission distance when the secret key rate is determined without a repeater and the highest secret key rate when the transmission distance is determined [86]. The curve of CV-QKD in different users is the relationship between the secret key rate and the distance corresponding to the different network capacities of RM-QAN. As can be seen from the figure, the secret key rate of all schemes decreases with the increase of transmission distance. Our scheme does not exceed the limit of PLOB under the same transmission distance. In addition, when the network capacity increases, the secret key rate decreases due to the gradual increase of optical circulator noise and frequency cross talk noise. The secret key rate will decrease obviously with the increase of network capacity, since the assumed eavesdropper Eve can obtain signal, resulting in the increase of maximum information that Eve can get. However, RM-QAN can still support the encoding of 128 users. It conforms to the concept of “the last kilometer” multi-access network.
2. Network Capacity
Network capacity is defined as how many users the network system can support and is mainly affected by the branch number of the N:1 splitter. The network capacity of RM-QAN is also affected by noise. The physical noise analysis in Section 3 shows that only optical circulator noise and frequency cross talk noise are related to network capacity . When network capacity is small, frequency cross talk noise is the main component of excess noise. However, when network capacity is large, frequency cross talk noise tends to be constant, and optical circulator noise is the main part of the noise. According to previously calibrated parameters, the relationship among network capacity, transmission distance, and secret key rate is shown in Fig. 3.
Figure 3.Relationship among network capacity, transmission distance, and secret key rate. Parameters are set as
In Fig. 3, the abscissa represents the transmission distance, the ordinate represents network capacity, and the color legend on the right of the graph represents the value of the secret key rate of each user, which decreases gradually from red to blue. Different network capacity scenarios are distinguished in Fig. 3 due to the varying transmittance attenuation caused by different splitting ratios, resulting in a direct effect on the secret key rate of users. As can be seen from Fig. 3, when network capacity is determined, the secret key rate decreases gradually with the increase of transmission distance. When the transmission distance is determined, the secret key rate decreases with the increase of network capacity. When users are added, the change of excess noise is small, and the secret key rate of each user remains almost constant. Certainly, if the detector bandwidth is not high enough, the number of users will be mainly limited by the detector bandwidth under the FDM scheme. If the number of users is larger, QLT requires higher output optical power to ensure that QNU achieves a certain output modulation variance, and we can use a high output power laser. It reflects that the RM quantum network scheme can accept a high network capacity without a performance penalty. In practical implementation, to achieve more accurate modulation variance adjustment, the light entering QNU and the total optical power will be larger, so network capacity will decrease. If network capacity is determined, the secret key rate of each user will be stable and not affected when more users access this network. In addition, our scheme needs only an optical circulator and a phase modulator (PM) to plug in a user, which has a low cost requirement for new users to access and a small impact on the secret key rate of each existing user. In conclusion, this scheme has high practicability.
C. Experiment Verification
1. Experimental Setup
The optical structure of RM-QAN used in the experiment is shown in Fig. 4. First, the light generated by the laser of QLT is divided into two parts by a beam splitter (BS) with a 2:1 ratio. The laser we use in the experiment has a very narrow linewidth, typically 100 Hz. Therefore, the interval between light before transmission and after transmission is less than the coherent length of the laser. On this basis, we realize homodyne detection, which means the beat frequency is zero. Light of higher power acts as the LO of the system. Light of lower power passes through a variable optical attenuator (VOA) to the optical circulator. The function of QLT’s VOA is to weaken light into appropriate optical power. Light is transmitted from port1 to port2 of the optical circulator and to the 30 km optical fiber spool. Afterward, the continuous wave laser is transmitted to a BS through the optical fiber spool. Then the light is evenly divided into eight QNUs by BS with a ratio of 8:1. Due to the limitation of modulated RF signal ports, we connected three QNUs in one experiment and verified the experimental possibility of eight users through multiple experiments. After arriving at the QNU, light is transmitted to a VOA through port2 to port3 of the optical circulator. The function of QNU’s VOA is to balance the optical power of each user. However, due to the manual adjustment of VOA, the optical power of each user cannot be completely identical, so there are differences in the signal-to-noise ratio (SNR) of each user in the experimental results. In practical implementation, the electronically controlled VOA can accurately control the output optical power of the QNU side. Light then enters a PM for QKD modulation, which is achieved using an arbitrary waveform generator (AWG). Different QNUs modulate information on different carrier frequencies, which can be distinguished on the spectrum diagram. QNU1, QNU2, and QNU3 in Fig. 4 select 10 MHz, 20 MHz, and 30 MHz bands, respectively. Afterward, the signal light enters the polarization controller (PC). The function of QNU’s PC is to adjust the polarization of each user, which can be eliminated after we use all polarization-maintaining fiber in practical implementation. After passing through the PC, it enters through port1 of the optical circulator, exits through port2, and returns to BS with an optical ratio of 8:1. At BS, the signal lights of the QNUs are converged into an optical fiber. After passing through the optical fiber spool again, the signal light enters from port2 of the optical circulator at the QLT and comes out from port3. Then the signal light reaches the PC of QLT for overall polarization adjustment. Finally, the signal light and LO light enter the integrated coherent receiver (ICR). After detection, the signal is collected in an oscilloscope. According to the Nyquist sampling theorem, the sampling frequency must be higher than two times the highest signal frequency. In this case, there is no signal spectrum aliasing, and the signal can be completely recovered. We adopt the no-switching scheme [26] and the reverse reconciliation scheme [24], which means that we use data obtained by QLT as a basis for reconciliation. In the case of reverse reconciliation, we can judge the maximum amount of information that Eve can obtain through the parameter estimation process, and then remove this part in privacy amplification to ensure the security of the final secret key. In this scenario, it can resist Eve’s splitting attack.
Figure 4.Optical configuration of the round-trip multi-band scheme. First, the light transmitted by the laser of QLT is divided into two parts by a beam splitter (BS) with a 2:1 ratio. The light of higher power acts as the local oscillator (LO) of the system. The light of lower power passes through a variable optical attenuator (VOA) to the optical circulator. Light is transmitted from port1 to port2 of the optical circulator and to the 30 km optical fiber spool. Afterward, a continuous wave is transmitted to a BS through the optical fiber spool. After arriving at the QNU, light is transmitted to a VOA through port2 to port3 of the optical circulator, and then enters a phase modulator (PM) for signal modulation. Afterward, the signal light enters the polarization controller (PC). After passing through the PC, it enters through port1 of the optical circulator, exits through port2, and returns to BS. After passing through the optical fiber spool again, the signal light enters from port2 of the optical circulator at the QLT and comes out from port3. Then the signal light reaches the PC of QLT. Finally, the signal light and LO light enter the integrated coherent receiver (ICR).
2. Signal Processing
To separate the data of different users, FDM is adopted in the experiment. In the modulation of QNU, the modulated signal is the waveform generated by the baseband signal and corresponding carrier frequency. The signal can be represented by Eq. (2). Then the AWG loads the signal to the PM for phase modulation. There are two advantages to operating like this. First, the signal formed in this way has an extremely strong ability to resist phase shift and noise. Second, the signal must be an orthogonal four-state CV-QKD signal after coherent demodulation. Their relative positions will not be changed, but only by the overall rotation, which is convenient for signal recovery. After coherent demodulation, the signal must be an orthogonal four-state CV-QKD signal. Their relative positions will not be changed, but will be rotated only as a whole.
In the demodulation of QLT, bandpass filtering is first employed for different frequency bands. Afterward, QLT determines whether there is a signal on each registered frequency band. After that, the first phase shift recovery is carried out to recover the optical phase. Since the obtained signal is adopted spectrum shifting, QLT uses coherent demodulation to restore the original signal of the baseband. It should be noted that the coherent demodulation of QLT needs to be used for and . The formula of coherent demodulation can be expressed as Eq. (3). QLT uses low pass filtering of the baseband frequency to get the baseband signal. After downsampling, the second phase shift recovery is carried out to rotate the signal as a whole. Then, the signal is determined by cross-correlation to complete frame synchronization. In frame synchronization, it is necessary to pay attention to the effect of mean value on cross-correlation. After that, QLT will do the parameter estimation to evaluate excess noise and further the information Eve can acquire, excluding this part in the post-processing. We also monitor the input optical power at both QNU and QLT to close the potential practical security loophole.
3. Experimental Results
The experimental results of RM-QAN are presented in the following. As can be seen from Fig. 5, the 10 MHz, 20 MHz, and 30 MHz frequency bands modulated by three QNUs can be clearly seen in the mixed spectrum obtained by QLT. There is no spectrum aliasing in these bands. Since the modulation scheme is DMCS, the signal constellation presents four states, which correspond to the different values of in Eq. (1). Different values of correspond to different colors. As shown in constellation diagrams of three QNUs in Fig. 6, the signal variance gradually decreases to the variance of shot noise with the reduction of SNR. It can also be seen from the constellation diagrams that the four states become more and more indistinguishable with the decrease in SNR. However, the result of cross-correlation is still clear in Fig. 7, which can easily complete frame synchronization. Since AWG is a clock synchronized with the oscilloscope, each red point in Fig. 7 should have the same horizontal coordinate, which means the frame synchronization position should be with the same value. It was well verified in our experiments with excellent frame synchronization results. In addition, the excess noise of 100 frames of each QNU fluctuates around zero, as shown in Fig. 8. Excess noise mainly comes from frequency cross talk noise, and the fluctuation comes from the deviation of data statistics. Their average is slightly more than zero. Inconsistent noise from different users results from statistical deviation in finite samples. Since we filter out noise from other users, the noise from different users will tend to be same when the amount of data is large. In Fig. 9, the ordinate value corresponding to the dotted line is the achievable secret key rate under the condition of 30 km in our experiment, where the secret key rate of QNU1 is 825.82 bit/s, the secret key rate of QNU2 is 674.46 bit/s, and the secret key rate of QNU3 is 635.95 bit/s. At present, due to the limitation of modulated RF signal ports, we have carried out experimental verification with eight users in multiple experiments. The feasibility of realizing more users can be fully inferred from the results, which can be realized by a BS with multiple interfaces.
Figure 5.Spectrum diagram of the signal obtained by QLT through the coherent receiver. It describes the change of amplitude at different frequencies.
Figure 6.(a) Signal constellation of QNU1, with
Figure 7.Cross-correlation of the signal modulated by three QNUs and the signal received by QLT. It describes the change of the cross-correlation of three QNUs at different frames, where the red points represent the successful result.
Figure 8.Excess noise scatter diagram of three QNUs. It describes the change of the excess noise of three QNUs at different times, where the red line represents the mean value of excess noise of three QNUs. The mean excess noise of QNU1, QNU2, and QNU3 is 0.0054 SNU, 0.0040 SNU, and 0.0059 SNU, respectively.
Figure 9.Secret key rate curves of the three QNUs in the experiment. It describes the change of the secret key rate of three QNUs at different transmission distances, where the ordinate value corresponding to the dotted line is the secret key rate under the condition of 30 km achieved in our experiment. The secret key rate of QNU1 is 825.82 bit/s, the secret key rate of QNU2 is 674.46 bit/s, and the secret key rate of QNU3 is 635.95 bit/s.
3. DISCUSSION
A. Physical Noise Analysis
In RM-QAN, the total noise can be classified into trusted noises and untrusted noises according to the traditional QKD noise model. Trusted noises are those that can be calibrated by the receiver QLT, which cannot be controlled by Eve, and they constitute electronic noise in the secret key rate bound calculation model. Untrusted noises, however, are caused by channel non-idealities or device non-idealities. These noises cannot be calibrated accurately and are controllable by Eve. Therefore, these noises constitute excess noise in the secret key rate bound calculation model and indicate the extent of eavesdropping. To ensure the security of the QKD access network, we performed an analysis of actual physical noise. The untrusted noises introduced by our special scheme are Rayleigh backscattering noise , frequency cross talk noise , and optical circulator noise . These are all untrusted noises. At the same time, we also briefly analyze the classical untrusted noise and classical trusted noise in general CV-QKD. In this way, the system performance can be analyzed by this updated physical noise model.
1. Rayleigh Backscattering Noise
In our scheme, light is transmitted from QLT by a round-trip structure, modulated by QNU, and received by QLT. This round-trip structure has a scattering effect, resulting in noise. Scattering noise is mainly divided into Rayleigh backscattering noise and Raman scattering noise. Raman scattering noise mainly affects the wavelength division multiplexing (WDM) system [87]. However, our FDM access network uses a single laser, which makes the wavelength unique. Moreover, theoretical analysis shows that the influence of Raman scattering on CV-QKD can be ignored because the LO light acts as a filter. In conclusion, only Rayleigh backscattering noise needs to be considered [88].
Rayleigh backscattering noise is caused by the interference of noise photons in the same spectral segment in the coherent detection of QLT, which can neither be filtered out nor attenuated. Rayleigh backscattering can occur anywhere in the fiber, and it cannot be monitored by a timing detector. Therefore, Rayleigh backscattering noise is something that needs to be considered in our scheme. The number of scattered photons produced by the Rayleigh backscattering effect is [88]
As shown in Fig. 10, this image describes the relationship between Rayleigh backscattering noise and transmission distance . As can be seen from the figure, Rayleigh backscattering noise increases gradually with the increase of transmission distance, but its order of magnitude is always within an acceptable range.
Figure 10.Relationship between Rayleigh backscattering noise and transmission distance. Parameters are set as
2. Frequency Cross Talk Noise
Because our access network scheme uses FDM to distinguish users, it inevitably produces photon cross talk between different frequency bands. When filtering the frequency band of a single user, the photons from other frequency bands leak in and produce noise. This noise is called frequency cross talk noise .
Before considering the frequency cross talk noise of all other users to a single user, we can simply consider the noise between any two bands . First, frequency interval is an important parameter to describe noise between frequency bands . In addition, the intensity of signal light also affects interband noise , which can be described by modulation variance . As the FDM scheme needs to use a bandpass filter to separate signals in different bands, the parameters of the filter also affect the noise between bands . The influence of interband noise is also different for different filter types. The classical Butterworth filter, which is used in our experiment, will be analyzed in the following. The Butterworth filter is mainly affected by the passband range and stopband range of the bandpass filter. It should be noted that other parameters of the filter, such as sampling rate, maximum passband attenuation, and maximum stopband attenuation, can be taken as reasonable values. Within reasonable limits, these parameters have little influence on filtering results. In conclusion, interband noise is mainly affected by frequency interval , modulation variance , passband range, and stopband range.
Before establishing the QKD link, we calibrated the appropriate stopband and passband for the bandwidth filters of individual users through experimental test data in advance. By searching for the optimal SNR, the optimal passband and stopband are found. Naturally, it is also possible to determine a better adaptive dynamic filter through machine learning and other algorithms to achieve a better filtering effect, but this is not within the scope of this paper. After hundreds of experiments, we got a reasonable passband and stopband range. Since signals mainly exist in the first main lobe, we take the passband boundary as the first main lobe of the signal and the stopband boundary as the second main lobe of the signal. This filtering result may not be the best, but it has met our experimental requirements.
Since the filtering process is solving linear differential equations with constant coefficients, we use the Monte Carlo method to obtain the noise between frequency bands . After several simulations, we obtained the relation between frequency interval and frequency division noise under different modulation variances , as shown in Fig. 11.
Figure 11.Relation between frequency interval and frequency division noise under different modulation variances. Parameters are set as
The dots in the figure represent values obtained by the Monte Carlo method, and the curves are obtained by the nonlinear fitting method. It can be seen from the figure that the noise between frequency bands is proportional to the modulation variance and decreases with the increase of frequency interval . It is worth mentioning that is rounded to better represent it in Fig. 11. The value of does not affect the generality of the resulting formula. Thus, interband noise can be expressed as
Figure 12.Relation between network capacity and frequency cross talk noise. Parameters are set as
The dots in the figure represent discrete values of noise under different network capacities, and the curves represent continuous values derived from the nonlinear fitting method. As can be seen from the figure, the frequency cross talk noise influenced by all other users increases gradually with the increase of network capacity, but the increase rate decreases gradually. It can be described as
3. Optical Circulator Noise
The round-trip QAN introduces an optical circulator to complete the optical round-trip scheme. However, the optical circulator is not found in the classical CV-QKD system. Moreover, the noise introduced by the optical circulator is non-ignorable when there are a large number of users. Therefore, we analyze the noise effect of the optical circulator on the system.
An optical circulator is a multiport nonreciprocal optical device. Its function is to make the optical signal be transmitted only along the specified port order. The three-port optical circulator we use is transmitted from port1 to port2 and from port2 to port3. The unique parameters in the optical circulator are isolation and directionality. Directionality is also called the cross talk of the optical circulator. The isolation is due to the imperfection of the optical circulator. When light is reversed, some light will still pass through. Directionality is caused by the structure of the optical circulator and other reasons, so that part of the light passing through port1 is directly output from port3. The isolation of the optical circulator is defined as the ratio of input optical power to output optical power in reverse light transmission [89]. The directionality of the optical circulator is the ratio of the input optical power of port1 to the output optical power of port3 when port2 is terminated and there is no reflection [89]. Therefore, the isolation and directionality of the optical circulator can be expressed as
For different optical structures, the noise introduced by the optical circulator is different. For our scheme, as shown in Fig. 13, the following analysis can be carried out.
Figure 13.Schematic diagram of optical circulator noise. The optical circulator of QLT receives light from the laser at port1 and transmits it to QNU at port2. The signal light returned by QNU is then received at port2 and output to the detector at port3. The optical circulator of QNU receives light from the QLT at port2, and then transmits it to the PM at port3.
First, we analyze the optical circulator of QLT. It receives light from the laser at port1 and transmits it to QNU at port2. The signal light returned by QNU is then received at port2 and outputs to the detector at port3. As the signal light is returned to the optical circulator, some light will enter port1 from port2 and return to the laser. Therefore, we added an optical isolator behind the laser to avoid damaging the laser and generating noise. Lasers usually need optical isolators to protect them, so the optical isolator in our scheme is not shown in Fig. 13. For the light received by port1, some will be directly output to the detector from port3, which will become noise photons and generate noise. The noise can be described as
Second, we analyze the optical circulator of QNU. The optical circulator in the QNU receives light from the QLT at port2, and then transmits it to the PM at port3. Afterward, it receives the modulated signal light at port1 and outputs it to the QLT at port2. As light from the QLT enters the optical circulator, some enters port1 from port2, which finally enters the PM. However, light will be screened out by PM due to the unipolarity of PM and will not produce noise. For the signal light received by port1, some light will directly output to the PM from port3, becoming noise photons and generating noise. The noise can be described as
In summary, the total noise introduced by the optical circulator is
Based on the above analysis, assuming that , , , and , we can get the relation between optical circulator noise and transmission distance, and the relation between optical circulator noise and network capacity, as shown in Fig. 14.
Figure 14.(a) Relation between optical circulator noise and transmission distance. The image describes the change of optical circulator noise at different transmission distances. (b) Relation between optical circulator noise and network capacity. It describes the change of optical circulator noise at different network capacities. Parameters are set as
As seen from the figure, the noise of the optical circulator increases gradually with the increase of transmission distance and network capacity, but it is within the acceptable range.
4. Other Untrusted Noise
In addition to the above mentioned Rayleigh backscattering noise , frequency cross talk noise , and optical circulator noise , which are the characteristics of the scheme, untrusted noise also includes modulation noise [90], amplitude noise [90], and phase noise [90].
The first is modulation noise. Because of the uncertainty of modulation voltage, noise will be introduced in the modulation process, and we need to find the relationship between the macro uncertainty of voltage and the optical quadrature components. In Ref. [90], the relationship between modulation voltage and optical quadrature components has been derived, which can be expressed as
Because the LO used in coherent detection is physically realized by a laser, the laser has intensity noise and phase noise. The physical cause of the relative intensity noise of the laser is that the laser source uses the principle of excited radiation to produce more photons. In addition, the change in the number of photons caused by spontaneous radiation is reflected in the amplitude of the laser, forming amplitude noise. Therefore, amplitude noise can be calculated as
Consider that the typical parameters of the laser are , , , and amplitude noise .
The spontaneous radiation of the laser causes not only the change of intensity but also the random change in the frequency of the laser pulse signal, forming phase noise. Due to the spontaneous radiation phenomenon inside the semiconductor laser, the photon generated by it is random in polarization and phase, which will directly affect the amplitude and phase of the light field formed by the excited radiation. For amplitudes, the magnitude is restored to the mean by glazing the radiation field with the inversion of the particle number in the laser medium, but the phase has no such resilience. Therefore, phase noise can be described as
According to the conventional parameter of coherence detection to weak coherent light, assume that , , , and phase noise .
5. Trusted Noise
In addition to the untrusted noise mentioned above, there is also trusted noise that can be calibrated or controlled by QLT. Here we briefly describe three kinds of common trusted noise: detector thermal noise , analog-to-digital converter (ADC) quantization noise , and common-mode rejection ratio (CMRR) noise [90].
Detector thermal noise is a kind of white Gaussian noise. Its amplitude distribution is Gaussian, the mathematical expectation is zero, and its power spectral density is constant. Thermal noise is a kind of noise produced by electronic components in the system, mainly by resistors and metal–oxide–semiconductor (MOS) tubes. The generation of resistance thermal noise is related to the thermal motion of electrons. Therefore, the thermal noise of the detector can be expressed as
The weak coherent state is detected and amplified by a balanced detector such that the final output voltage signal is proportional to the measured canonical component. However, if the output voltage is quantized by an ADC, the ADC introduces additional noise to the weak coherent state, making the excess noise larger. The same as detector noise, we convert a macroscopic physical quantity such as ADC noise into noise on the canonical component of the quantum state. Therefore, ADC quantization noise can be expressed as
We can estimate ADC noise by common parameters , , , , , , , and ; then .
A practical differential amplifier in a balanced detector amplifies not only the differential currents, but also their average currents. If a heterodyne detector consisting of two homodyne detectors is used for detection, the final noise introduced by CMRR is
Assuming that , , , , , , and , in this case, . This noise depends largely on the size of the relative intensity noise of the LO, so the relative intensity noise of the LO is required to be extremely low.
4. CONCLUSION
In conclusion, we propose a flexible and efficient quantum network physical structure, namely, RM-QAN. In detail, this QAN can make quantum states travel in a circle to transmit data due to the round-trip structure. It can also support multi-user access through multi-band quantum state transmission and separation. Based on this proposed network, we realize multi-user secure key sharing through CV-QKD. The theoretical noise model of the multi-user CV-QKD has been established, the user capacity and theoretical key rate of the scheme have been discussed, and the proof-of-principle experimental verification has been carried out. The proof-of-principle experiment shows that each QNU can share a practical secret key rate of about 600 bit/s at a transmission distance of 30 km with QLT.
Certainly, there are three main challenges in the practical implementation of this scheme. (1) At this stage, we analyze the secret key rate bounds in the asymptotic regime, and the finite-size effects and composable security are not considered, which are critical to the final practical implementation. In such a security framework, we need longer data blocks to overcome the finite-size effect, which requires a higher communication rate and larger data storage, and therefore has requirements for hardware and algorithms. (2) The practical access of more users also needs a classical communication protocol related to the data link layer and network layer to realize information synchronization, such as the handshake protocol between QNU and QLT. After accessing more QNUs, QLT needs to exchange more classical data in the post-processing stage, requiring the high bandwidth of the classical channel. (3) For practical security, the effective countermeasures of the phase remapping attack [80,81] and the Trojan-horse attack [79] have been proposed and can be added in further implementation.
The advantage of our scheme is the simple optical structure in the physical layer. Specifically, the round-trip structure requires only one laser and one detector in the entire network, and only one modulator and one circulator need to be plugged in when a new user accesses. Performance evaluation shows such networks have low physical excess noise for each user theoretically and can support multi-user access and quantum secure key generation. Moreover, this scheme can coexist with classical communication, since classical communication can adopt different frequency bands in FDM. Combined with the comprehensive protocols of the data link layer and network layer and practical security countermeasures, this scheme can be an effective solution for QKD access network establishment. This work lays the foundation for the subsequent establishment and application of a large-scale and multi-user QAN.
APPENDIX A
In the following, we derive the secret key rate of DMCS CV-QKD under collective attack. Specifically, in the case of reverse negotiation, the secret key rate for unit system repetition rate can be written as
The core of secret key rate calculation is to evaluate the upper bound of the information Eve steals. Under collective attack, the Holevo bound is used to limit the maximum information Eve can extract from Bob, so is
Theoretical security analysis of the CV-QKD protocol under collective attack shows that under the condition of known covariance matrix of state , if Eve’s eavesdropping operation is a Gaussian operation, it can get the most information, which is called “Gaussian attack optimality theorem.” The theorem states that if the final quantum state shared by Alice and Bob is regarded as a Gaussian state, the calculated stolen information by Eve is an upper bound of the real stolen information. The information entropy calculation of the Gaussian state is relatively simple, which means the above equation can be simplified as
It can be found that this matrix is similar to the covariance matrix in the Gaussian modulation coherent state (GMCS) CV-QKD protocol, except that the Einstein–Podolsky–Rosen (EPR) correlation in the GMCS CV-QKD protocol is . When , is extremely close to . Under this condition, it can be considered that the information Eve steals from Bob is equal in both protocols. Based on this conclusion, the secret key rate can be deduced according to the GMCS CV-QKD protocol, and corresponding parameters and are
On the other hand, matrix can be calculated as
After obtaining the above matrix, we can find the symplectic eigenvalues of matrix . In the following, we directly give their calculation formula:
The last symplectic eigenvalue is
References
[1] S. Khatri. Towards a general framework for practical quantum network protocols(2021).
[2] H. J. Kimble. The quantum internet. Nature, 453, 1023-1030(2008).
[13] C. Simon. Towards a global quantum network. Nat. Photonics, 11, 678-680(2017).
[43] . OIDA Quantum Photonics Roadmap(2020).
[78] https://www.idquantique.com. https://www.idquantique.com
[89] R. Hui. Introduction to Fiber-Optic Communications(2019).
Set citation alerts for the article
Please enter your email address