Quantum secure direct communication (QSDC) is a secure communication paradigm that transmits information directly using quantum states [1, 2]. Leveraging the fundamental principles of quantum physics, it inherently provides eavesdropping detection capabilities and provable information-theoretic security. Owing to its weak optical signal transmission, QSDC also exhibits covert communication capabilities. It inherits the advantages of optical communication, such as license-exempt operation and resistance to electromagnetic interference. Various experiments have also demonstrated this elegant and robust paradigm for transmitting classical information using quantum states [3,4,5,6,7,8]. It is gradually demonstrating its potential for large-scale quantum network deployment [9,10,11,12,13,14,15,16].

Communication rate, communication distance, reliability, security, and communication overhead are key performance metrics of QSDC. From the security perspective, the introduction of decoy states enables the detection of photon-number-splitting attacks [17, 18], while removing specific active modulation avoids potential side-channel vulnerabilities [19, 20], thereby enhancing both the secrecy capacity and the secure communication distance of QSDC. If the measurements in QSDC are performed by an untrusted third party, all detector side-channel attacks can be removed [21,22,23], and an equivalent level of security can also be achieved through simplified implementations [24, 25]. Moreover, communication security can be guaranteed solely by observing violations of Bell inequalities, without requiring any detailed description of, or trust in, the internal working principles of the communication devices [26, 27]. Classical forward error-correction code [9] and quantum error-correction code [28] can further improve the reliable transmission of messages encoded in fragile quantum states, with classical techniques being particularly mature and well established [9].

However, due to its weak optical nature—for instance, requiring single-photon emission [9] and the consequent prohibition of signal amplification—the achievable transmission rate and efficiency of current QSDC systems are limited. At present, constrained by device performance and channel losses, the transmission rate over distances on the order of a hundred kilometers remains at the kilobits-per-second level [9]. Within distances of up to several tens of kilometers, the infinite-dimensional and continuous encoding space of continuous-variable quantum states is beneficial for achieving higher transmission rates [29,30,31]. Nevertheless, simultaneously improving both the transmission rate and the achievable communication distance of QSDC remains a significant challenge.

Semantic communication, as a next-generation communication paradigm, transcends the traditional Shannon information framework for bit-level transmission by focusing on the effective delivery of semantic information [32]. Leveraging semantic modeling and deep learning techniques, semantic communication can compress redundant information and achieve improved effective transmission rates under limited channel resources, providing a promising approach to mitigate the rate bottleneck of QSDC.

Recent advances in semantic communication have demonstrated enhanced security through private models [33], where the knowledge base and semantic models are kept private to reduce the probability of decoding by potential adversaries. Nonetheless, semantic communication alone still faces security challenges under sophisticated eavesdropping or tampering attacks. Through the integration of semantic communication with QSDC, semantic information can be transmitted at the physical layer via quantum states, simultaneously ensuring high transmission efficiency and further strengthening security. This integration inherits the physical-layer eavesdropping immunity of QSDC while leveraging the advantages in transmission rate and efficiency of semantic communication, laying the foundation for building secure and high-throughput semantic quantum networks [10].

Here, we propose the semantic QSDC paradigm. A semantic QSDC process contains two stages. During the training stage, Alice and Bob construct the knowledge base by training. Then they go to the communication stage, where Alice uses the knowledge base to compress the information into semantic information, and then sends it to Bob through a QSDC protocol. Bob receives the semantic information and decode it with the help of the knowledge base. In the following, two types of semantic QSDC protocols are proposed, one is based on separated designs, and the other based on joint designs.

The incipient integration of quantum theory and semantics mainly draws on mathematical and conceptual frameworks from quantum mechanics, such as density matrices [34], superposition, measurement-induced collapse, entanglement [35, 36], and contextuality, to model the correlations [37], dynamics, uncertainty, and observer dependence of meaning in natural language [38]. Quantum resources have also been exploited to enhance semantic extraction [39, 40], while semantic features have in turn been used to improve the robustness of quantum communication [41]. In addition, semantic communication concepts have been introduced into single-photon LiDAR systems, where target recognition is formulated as a semantic communication task through a self-updating semantic knowledge base, and semantic feature encoding improves noise tolerance and adaptability [42].

Furthermore, incorporating semantic security can provide stronger and more practical confidentiality than conventional strong security [43,44,45,46]. Strong security only guarantees that an eavesdropper cannot obtain information when messages are uniformly distributed, whereas semantic security requires that no useful information can be extracted by an eavesdropper regardless of the message distribution, including non-uniform distributions that may be known to or even chosen by the eavesdropper. These advances are beneficial for strengthening the security proofs of QSDC.

In the classical domain, semantic communication has already been optimized and applied in practical optical communication systems [47,48,49]. By contrast, the term “quantum semantic communication” has appeared in a number of proposed schemes. The semantic extraction employs quantum embedding techniques to map classical data into high-dimensional Hilbert spaces and generate quantum states, and quantum clustering algorithms are then used to extract contextual and semantic features, yielding semantic centroid vectors. During the encoding stage, orbital angular momentum [50, 51] is employed to generate transmissible quantum representations in the form of qudits. During transmission, entanglement purification protocols are applied to mitigate channel noise. At the measurement stage, the receiver performs projective measurements on the quantum states and, combined with quantum gate operations, recovers the embedded semantic information. Semantic reconstruction is then achieved by exploiting the mapping between semantic centroid vectors and quantum states, enabling accurate interpretation from quantum states to semantic concepts [50, 51]. Quantum computing algorithms have also been proposed to encode classical data from knowledge graphs into quantum states [52, 53], with transmission further realized via quantum remote state preparation [54]. Alternatively, classical semantic information bits 0 and 1 has been directly mapped onto corresponding quantum states \(|0\rangle\) and \(|1\rangle\) for transmission [55, 56]. However, such approaches may suffer from security vulnerabilities because orthogonal-basis quantum-state encoding is not adequately considered. To enhance the security of semantic information transmission, schemes based on quantum anonymous broadcast have been proposed [57]. High-performance communication applications [58] and candidate communication paradigms for 6G networks [59,60,61] have also begun to envision these concepts. Nevertheless, these approaches still face substantial challenges in semantic extraction, practical communication implementation, and security enhancement, as they rely on quantum computing and quantum communication protocols that have yet to be fully realized in practice.

3.1 A separated-design-based semantic QSDC protocol

The separated design semantic QSDC protocol is illustrated in Fig. 1. The semantic QSDC system based on a separated design follows the steps below.

• Step 1, semantic encoding. Based on a semantic knowledge base, the sender Alice performs semantic encoding of the source information and outputs semantic information. The key to semantic extraction lies in employing appropriate semantic models that utilize artificial intelligence, particularly deep learning, to extract higher-level semantic information from the source data.

• Step 2, source encoding. The semantic information is further processed by the source encoder to yield source codewords.

• Step 3, forward error correction coding. The source codewords are mapped into transmitted codewords via a forward error correction encoder. Typical options include low-density parity-check (LDPC) codes and polar codes.

• Step 4, quantum state modulation and transmission. The QSDC transmitter modulates the codewords directly onto quantum states according to a chosen QSDC scheme, and delivers them to the receiver Bob through a quantum channel. Depending on the deployment, the channel may be an optical fiber link or a free-space optical link [17, 62].

• Step 5, quantum state measurement. Upon reception, the quantum states are detected by Bob to obtain the received codewords.

• Step 6, forward error correction decoding. Bob decodes the received codewords using the forward error correction decoder, recovering the source codewords.

• Step 7, source decoding. The source codewords are passed through the source decoder to reconstruct the semantic information.

• Step 8, semantic decoding. The receiver decodes the semantic information based on the semantic knowledge base and a semantic decoding model to reconstruct the original message.

• Step 9, dynamic knowledge base update. To adapt to evolving environments and application requirements, and to ensure timeliness, accuracy, and relevance of communication, the semantic knowledge base shared between the transmitter and receiver must be dynamically updated.

The semantic QSDC system based on a separated-design framework is under a classical–quantum hybrid architecture. The system cascades classical semantic communication coding with QSDC, requiring only the addition of semantic encoding and decoding modules to an existing QSDC system. The semantic, source, and channel coding modules are designed independently, ensuring strong compatibility with conventional QSDC architectures and resulting in low modification costs for practical deployment. The only additional required interface is a classical digital interface between the semantic module and the QSDC encoder/decoder, such as a binary sequence or symbol stream.

It is worth noting that such a separated design can be further modified by integrating source coding and forward error correction coding into a unified semantic encoder, jointly trained with respect to quantum channel characteristics. We refer to the above system as a “separated design” because it represents a classical–quantum hybrid paradigm, where conventional information encoding/decoding is combined with QSDC transmission.

3.2 A joint-design-based semantic QSDC protocol

The semantic QSDC based on a joint design is illustrated in Fig. 2. It is operated through the following steps:

• Step 1, semantic encoding and signal generation. The transmitter Alice, leveraging both the semantic knowledge base and the source information, produces quantum signals through an semantic QSDC transmitter. This transmitter is realized as a neural network jointly designed for semantic encoding, source–channel coding, and quantum signal modulation. It directly drives the optical modulation circuit to generate quantum signals to be transmitted, thereby embedding semantic-level information into quantum states.

• Step 2, quantum signal transmission. The quantum signals are transmitted to the receiver through either an optical fiber or a free-space quantum channel.

• Step 3, semantic decoding. Upon reception, the receiver Bob reconstructs the transmitted information using the semantic knowledge base and a semantic QSDC receiver. The structure of the semantic QSDC receiver corresponds to that of the transmitter, employing an integrated neural network jointly designed for quantum signal demodulation and channel–source–semantic decoding.

• Step 4, dynamic knowledge base update. To maintain adaptability to varying environments and application requirements while ensuring the timeliness, accuracy, and relevance of transmitted information, the semantic knowledge base shared by the transmitter and receiver should be continuously updated. When channel conditions vary, the communication terminals at both ends must retrain their neural networks to maintain reliable semantic QSDC.

By employing an end-to-end training approach, the above system not only effectively extracts semantic features from the source data but also accounts for the characteristics of the quantum channel, enabling a joint design of source, channel, semantic encoding, signal generation, and their combined reverse process. This approach enhances the system’s adaptability to channel conditions, thereby significantly improving overall performance. However, the structural form of this design remains an issue worthy of further investigation.

Before deploying the two aforementioned schemes, both the transmitter and the receiver employ an identical pre-trained model with fixed parameters as their initial semantic knowledge base. This model has already been trained on large-scale, general-purpose datasets. The knowledge base can be pre-shared, updated online, locally trained, or updated offline. Joint online learning and incremental updates can then be carried out through either classical communication or QSDC. When QSDC is used, the size of the transmitted content must be matched to the available throughput, and at the current stage it is practical to transmit only a small amount of critical information for knowledge base synchronization at both ends.

Semantic communication typically assumes trusted devices, meaning that privatizing the knowledge base can further enhance transmission security. Furthermore, transmitting semantic information via QSDC can mitigate security risks associated with knowledge-base leakage, as an eavesdropper cannot reliably obtain the transmitted content from a quantum channel. The separate design paradigm offers significant advantages in terms of lower complexity, module independence, and ease of implementation and deployment, making it the cornerstone of contemporary large-scale and heterogeneous networks. In contrast, the joint design paradigm has superior potential for higher efficiency and performance, positioning it as a key enabling technology for future communication systems with stringent demands, such as high efficiency and low latency. A distinctive strength of this approach is its seamless incorporation of signal modulation and demodulation optimization into the dynamic training process of neural networks. However, this integrated optimization inevitably comes at the cost of significantly increased system complexity.

Our proposed semantic QSDC schemes differ from previous quantum semantic communication approaches [50,51,52,53,54,55,56,57,58,59,60,61] in several significant aspects. First, we adopt a hybrid classical–quantum architecture in which semantic encoding and decoding rely on mature classical techniques and do not require quantum computing support [50,51,52,53, 57, 58, 61], thereby substantially reducing technical complexity. Second, secure transmission of semantic information is realized through a QSDC protocol, which exploits intrinsic quantum properties to provide strong security guarantees such as eavesdropping detection, and whose underlying technologies have already reached a practical level of maturity. Third, the proposed scheme focuses on security enhancement at the transmission layer and on improving quantum communication efficiency, achieving a clear decoupling between security mechanisms and semantic processing. This design simultaneously ensures high security and preserves the flexibility of classical semantic solutions, resulting in excellent mutual compatibility. Finally, the joint design paradigm points toward a promising direction for future cross-layer, cross-module, and end-to-end collaborative optimization.

To evaluate the performance of semantic QSDC, we consider a separated-design scheme where semantic encoding is integrated with a one-way quasi-QSDC system [9] in simulations. The secrecy capacity of the semantic QSDC employing a weak coherent pulse source can be expressed as

In conclusion, semantic QSDC represents a novel paradigm that integrates the principles of semantic communication with QSDC, simultaneously achieving high transmission efficiency and enhanced security. By directly embedding semantic-level information into quantum states, semantic QSDC not only overcomes the rate limitations of conventional QSDC systems but also utilizes semantic modeling to optimize both the transmission and decoding processes. Remarkably, semantic QSDC possesses the capability to surpass the Shannonian mutual information bound and break the linear bound of quantum communication. This breakthrough does not require sophisticated physical-layer architecture designs. It should be noted that the separation-based semantic QSDC scheme is universal, representing a hybrid classical–quantum architecture. This paradigm is compatible with all theoretical QSDC protocols and can be rapidly implemented experimentally without any modifications to the communication apparatus. Our approach opens new avenues for intelligent, secure, high-speed, and task-oriented quantum communication networks. Furthermore, QSDC provides inherent security guarantees for semantic communication, significantly enhancing its resistance to eavesdropping attacks.