- Open Access
Development and validation of a safe communication protocol compliant to railway standards
Journal of the Brazilian Computer Society volume 27, Article number: 5 (2021)
Railway systems are composed of a multitude of subsystems, sensors, and actuators that exchange datagrams through safety-critical communication protocols. However, the vast majority of these protocols rely on ad hoc interlacing mechanisms and safety codes which raise the heterogeneity and complexity of the overarching railway system. Therefore, Rete Ferroviaria Italiana, the company who is in charge of managing the Italian railway network, coordinated the definition of the Protocollo Vitale Standard (Standard Vital Protocol). This protocol is inspired to, and compliant with, the communication protocols adopted for the European Train Control System (ETCS) (SUBSET, UNISIG, 037, Euroradio FIS, version 2.3. 0; SUBSET, UNISIG, 098, RBC-RBC safe communication interface, 2007), and it is meant to become the standard layer to enable safe communication between components of the Italian railway system. This paper reports our experience in the design, implementation, verification, and validation of the Protocollo Vitale Standard in compliance with the European safety standards for railway systems. We first defined a safety plan and a verification and validation plan, which guide the design, development, verification, and validation activities as required by safety standards. Guidelines of such plans have been followed strictly until completion of the work, which concludes with the provision of a safety case where all safety evidences are summarized. Noticeably, we (i) selected appropriate safety mechanisms, (ii) verified the software design, (iii) implemented the software in compliance with code metrics and coding rules, (iv) conducted tests to validate the protocol against its functional and performance requirements, and ultimately (v) devised all relevant documentation and a safety case which summarizes the evidences needed for certification.
Safety-critical systems must adhere to appropriate guidelines to ensure that safety requirements are met. In fact, following the definition of safety, i.e., avoidance of catastrophic failures [1, 2], any misbehavior shall not lead to fatalities, severe injuries, or major damages to the environment [3, 4]. In other words, a safety-critical system must be able to mitigate and manage potential catastrophic failures. In addition, security breaches or vulnerabilities could also lead to unsafe behaviors; therefore, safety-critical systems may also be required to guarantee security, often with a particular accent on integrity of information. Safety standards typically define a safety integrity level (SIL) [5, 6], that sets qualitative and quantitative constraints that must be met to ensure the safe behavior of a target component or system. Matching the requirements of a desired safety integrity levels mean that a specific set of processes and techniques have been applied through the lifecycle of the system, and such application is documented by a specific list of work products.
Safety standards for the railway domain
In the railway domain, different devices, actuators, or control systems need to cooperate with each other to provide safe functionalities. In order to be actually approved and deployed in Europe, railway systems and protocols must be designed, implemented, verified, and validated in compliance with the CENELEC standards. Noteworthy, safety of communication for electronic and electrotechnical equipment in the railway is regulated by the CENELEC EN-50159 . In addition, standards CENELEC EN-50126 , EN-50128 , and EN-50129  describe the required processes and techniques for the entire lifecycle of system, software, and hardware.
Interlacing and communications
When these devices are placed in different locations, they must communicate remotely through appropriate transmission systems such as LAN, WAN, or GSM-R . The data sent over the transmission system is often critical, containing information that will be used to perform safety-related activities e.g., changing the state of a rail switch. Consequently, such components or subsystems cannot rely on basic protocols developed for the exchange of non-critical information as TCP/IP , as they do not offer sufficient safety guarantees. Protocols to be exercised in a safety-related environment should comply with the standards that are defined for the specific domain. Typical protocols that take part to the deployment of a railway system are Euroradio , used for the GSM-R transmission system in the European Train Control System (ETCS) level 1 , and the protocol which manages communications between Radio Block Centres (RBCs) , servers in charge of continuously transmitting to trains their speed limit and the movement authority.
This paper documents our experience in the design, implementation, verification, and validation of software for the railway domain. We show how to follow applicable standards, motivate key choices, and show several outcomes of V&V activities. The overall process is applied to the Protocollo Vitale Standard (PVS), a communication protocol that provides a safe interface between railway actuators, sensors, and control systems. The protocol was specified by Rete Ferroviaria Italiana (RFI, the company who manages the Italian railways) as a standard communication protocol built upon existing—albeit heterogeneous—protocols as [12, 14]. Our work is based on [6, 7] and follows the lifecycle described in the CENELEC standards, starting from the Software Requirements phase up to the step 6.B–Software Validation, but does not account for further steps, which are currently managed by the owner of the case study. In order to be applied into railway systems, PVS should conform to SIL4, which is the highest SIL defined by CENELEC standards. We devised a verification, validation (V&V) and safety plan that describes, among others, techniques for software architecture, design specification, implementation, and test specification and execution. This paper extends the work in , where we provided a preliminary explanation of activities performed during the project’s lifecycle. Instead, this work depicts the whole process and set of techniques and methodologies applied to develop the protocol. Relevant additions include (i) a revised architecture specification, (ii) a detailed dynamic analysis including the resulting coverage analysis and related discussion, (iii) conclusive results on static analysis and related discussion.
This paper is structured as follows: “Control systems and railway standards” section describes relevant systems in the railway domain, along with applicable standards. “Specification of Protocollo Vitale Standard” section describes the main design patterns, strengths, weaknesses, and range of applicability of the PVS. The “Verification, validation, and safety plan” section expands the V&V plan and the safety plan that describe the documents to be produced and the techniques selected for each phase. The application of such techniques is detailed in “Architecture specification” and “Design specification” sections, letting “Verification and validation activities” section report on code metrics and testing. Performance analyses are expanded in “Performance analysis of the protocol” section, while “Concluding remarks” section concludes the paper.
Control systems and railway standards
In this section, we report on control systems that are relevant for European railways, alongside with a summary of the CENELEC standards which provide prescriptive guidelines.
The realization of the European Railway Traffic Management System (ERTMS, 0) requires the deployment and interaction of heterogeneous devices, sensors, actuators and embedded systems, often with human interactions, e.g., the train driver. Among the different components, the Station Control Systems (SCSs) provide computer-based interlocking to (i) manage the interactions between trains and physical devices, e.g., semaphores, railway switches, as well as (ii) connecting the train with peripheral or central control systems. To avoid collisions, trains periodically ask the movement authority, which grants or denies authorization to move forward through their route. The Radio Block Centers (RBCs) instead are trackside units that interact with central offices and SCSs to provide feedbacks through the GSM-R network .
Currently, two ERTMS functional levels implement different communication strategy between trains and control systems. In ERTMS Level 1, trains communicate with the Balise Transmission System (BTM) , which relies on rail beacons called Eurobalises that measure speed and location and provide them as telegrams to trains as soon as they pass by. As it can be seen in Fig. 1, ERTMS level 2 is instead a digital radio-based system, where data are handled by the RBCs. With this configuration, eurobalises are considered reference points for correcting distance measurement errors. Since radio transmission systems are susceptible to a number of problems related to security aspects and integrity of data, ERTMS level 2 forces devices and subsystems to adopt the Euroradio FIS  (Euroradio Functional Interface Specification). This enables communications between critical components to be carried out in accordance to the safety standards specified by the Category 3 (open networks) of EN50159 .
However, while Euroradio FIS is a requirement for Over-The-Air communication, there are no unified standards to protect communications between other entities e.g., between actuators and interlocking. Therefore, each manufacturer devises its own protocol and deploys it wherever it is needed. The lack of a unique safe communication protocol was the pilot reason for the definition of the Protocollo Vitale Standard (PVS) specification, which generalizes the Euroradio protocols and is intended to be applied in all the communications in the Italian railway system. For example, PVS can act as safe and secure channel for the communication between RBCs and the SCS, or more in general to interlace various control/signalling systems produced by different manufacturers.
Safety-critical systems need to be designed, developed, and deployed in compliance with standards, which provide appropriate guidelines. Focusing on the railway domain, the CENELEC EN50126  standard defines the safety integrity levels (SILs) and the corresponding lifecycle for hardware-software systems. The standard defines activities to be performed through the lifecycle, ranging from System Definition to Operation and Maintenance. As a result, several documents are produced, such as (i) the hazard log, where all the possible situations that might lead to damage to the health of people, properties, or the environment (hazards) are listed alongside with countermeasures and mitigations; (ii) the safety requirements document, which reports actions to reduce the severity and the probability of occurrence (risk, ) associated to each hazards; (iii) the safety plan, which guides resources management, the activities, and the responsibilities involved to reach a given SIL, and (iv) the safety case, which provides evidence that actions required in the safety plan were correctly executed and safety measures are effective and in place.
When the system contains (or is) software, compliance should also be met with the EN50128 , which focuses on the development of software for railway control and protection systems. Among others, the standard drives the definition of a verification and validation plan (V&V Plan), which specifies all the activities prior, during and after the design and development of the system. EN50128 expands on a wide set of techniques to be applied during the development lifecycle, which can be selected according to the desired SIL.
In addition, also interactions and communications between system components should be strictly regulated to avoid unsafe situations. To such extent, EN50159  describes how to manage and protect safety-related communications depending on the underlying transmission system and its characteristics. EN50159 identifies accidental and malicious threats to safety-related communications and recommend possible countermeasures.
Specification of Protocollo Vitale Standard
The Protocollo Vitale Standard (PVS) is a communication protocol that lays over the ISO/OSI Transport and implements the Session and Presentation ISO/OSI levels . Both TCP and UDP transport mechanisms are supported by this protocol, which defines slight variations of its structure to perfectly suit either of the two. Overall, PVS is a light protocol that can wrap communications between (i) Station Control Systems (SCSs), (ii) SCSs and RBCs, and (iii) SCSs and Graphical Elaboration Assembly (GEA). While protocols [12, 14] are tailored for their respective domains, they may not be portable to a slightly different context. As an additional merit, PVS can deal with both Category 1 (closed transmission system, e.g., private LAN) and Category 3 (open transmission system, e.g., public Wi-Fi) transmission systems as specified in .
The reference architecture of the PVS (see Fig. 2) is designed  as a stack composed by three mandatory layers, plus an optional one used to provide additional cryptographic functionalities. In this way, the PVS can deal with both Category 1 (closed network) and 3 (open network) transmission systems, providing defences against (i) data corruption, (ii) resequencing, repetition, insertion, delay, and deletion of packets, and (iii) masquerade in compliance with EN50159 . Each layer has different responsibilities:
Safe application interface layer (SAI): manages the data flow between the application and other PVS layers by applying several defense mechanisms, namely (i) sequence number, that protects against resequencing, repetition, insertion, and deletion, and (ii) execution cycle, that ensures “freshness” to shield against delay attacks.
Safety layer (SL): placed between SAI and APL layers, it ensures message integrity through non-cryptographic safety code and authenticity by means of a node identifier, called nSAcePID.
Adaptation and redundancy management layer entity (ALE): manages redundancy (for availability purposes) by operating on two physical communication links and guarantees correct interfacing with the underlying transport mechanism, either TCP or UDP.
Access protection layer (APL–optional): required only when dealing with Category 3 (open) transmission systems; it provides access protection through AES  and AES-CMAC  cryptographic algorithms.
The communication is cyclic with configurable period, which is set independently for each device: typically, railway devices range from 250 to 500 ms. Each producer must periodically send the information according to its own cycle time, which will be processed by the consumer. Symmetrically, each consumer must acquire cyclically, according to a period independent of the producer’s cycle, the information sent by the producer. When an application wants to send user_data, the bytes flow goes through the protocol stack in Fig. 3 where each layer wraps the message by adding its header/footer information. When receiving, each layer in reverse order decapsulates the message by removing either header/footer and then forwards the unwrapped data to the upper layer.
When the communication does not need cryptography, APL can be omitted. Moreover, in Fig. 2 ALE is depicted externally of the “safety related equipment” area meaning that this layer can be deployed on a remote (non-safe) device because it does not have safety requirements. For such reason, a means other than shared memory must be provided to realize the communication between ALE and the other layers of the protocol stack.
In terms of handling of the connection, the most important role is played by the safety layer entity: when establishing a connection, the safety layer activates entity authentication, a handshake where two devices exchange (i) a nonce, (ii) the initial sequence number, (iii) execution cycle, and (iv) device identifier. In Fig. 4, the initial state of the safety layer is marked as IDLE: here, the safety layer is waiting for connection establishment requests. When a request arrives and the handshake is performed correctly, the safety layer goes through WFAU3 and WFRESP in case of an incoming connection establishment request, through WFTC, WFAR whenever an outgoing connection request arrives. In both cases, a correct handshake leads the safety layer entity in the DATA state, where the two devices can finally exchange safety-related messages. The occurrence of one or more errors during the handshake or the exchange of data messages cause a transition to the IDLE state, which is therefore the safe state of the PVS.
Verification, validation, and safety plan
The first steps for the realization of a safe system according to the railway standards are the definition of a V&V and safety plan. The plan describes the steps to be fulfilled during the realization of the PVS to comply with applicable standards, to ultimately ensure a proper planning of activities which are adequate to reach the target SIL. Since the PVS is essentially software, we refer to the lifecycle of the CENELEC EN-50128 . We explored such standard to define the documentation plan, which enlists the various documents to be produced at different stages of the design, implementation, validation, and deployment of a component or a system. Table 1 reports the documentation plan tailored to PVS, highlighting the document name and number, as well as tracing the document with respect to the involved EN50128 activities. It is worth remarking that our list of documents is a subset of all the lifecycle documents described in  (page 66, Table A.1), as our activity focuses only on the software requirement specification, architecture specification, design and implementation and verification and validation, while requirements definition, integration, deployment, and decommissioning are left out.
Planning: description of the documentation plan, scheduling of the verification and validation activities and selection of the techniques or methodologies that must be adopted to obtain the target SIL during each of the development phases.
Software requirements: description of the software system to be developed considering both functional and non-functional requirements, development of the test plan to be followed during the verification phase.
Architecture and design: definition of the software architecture to specify the software elements composing the system, the relations among them, their properties and their tasks. Definition of the software design to clarify the implementation activity to the development team with respect to what defined in the software architecture specification document, definition of the software interface for the interaction with the system and verification of the system architecture and design with respect to the verification validation and safety plan.
Component design: description of the design of each architecture’s component, specification of the tests for each component and verification of the component design.
Component implementation and testing: implementation of the components, definition of the source code documentation, verification of the tests performed for each component, and of the source code produced.
Overall software testing/final validation: reporting on the overall test performed over the software system and final validation of all the activities performed during the system development.
For each of the phases above, except for the planning phase which requires the drafting of the safety and V&V plan itself, in the safety and V&V plan we defined the techniques adopted to manage the safety and reach the target SIL, described the methodologies to be followed during the phase execution and scheduled the activities needed to release the final product. In the following section, we will report on the details about each of those project’s phases, which, for simplicity, will be divided on “Architecture specification” section, “Design specification” section, and “Verification and validation activities” section.
The standard  suggests methodologies and techniques to be applied for architecture specification, depending on the target SIL. PVS needs to comply with SIL4, therefore, according to Table A.1 of  we chose to adopt defensive programming, error detecting codes, fully defined interfaces, structured methodology, and modeling, which are summarized in Table 2. Software architecture was specified through the ISO/IEC 10746-1 Open Distributed Processing (ODP-RM) , a structured methodology which requires the use of five different viewpoints, each of them covering different aspects such as the definition of the general SW architecture, the data-flow between the SW modules, the specification of the Protocol Data Unit(s) (PDUs), and interfaces.
Modeling techniques allow describing ODP-RM through (i) structure diagrams for the description of the SW architecture, (ii) a state-transition diagram to describe the behavior of the protocol, and (iii) sequence diagrams to describe the sequence of actions taken during the communication between two devices. We will expand briefly each of the ODP-RM points of view in “Enterprise viewpoint” section to “Technology viewpoint” section. In this way, we will also debate on the items of Table 2 that are yet to be discussed: defensive programming, error detecting codes, and fully defined interfaces.
Enterprise viewpoint specifies the role and responsibilities of (i) actors involved in the system and (ii) the external environment. It explains the relations between the system’s component and the internal and external interfaces. We hereby expand the general PVS architecture depicted in Fig. 5, which builds on the following entities:
Monitor: according to the Software Architecture/Design Specification (documents 9 and 10 in Table 1) this module checks the correct evolution of the control flow to provide defensive programming by implementing control flow monitoring, as it can be seen in Table 2;
Layer: the module that implements features shared amongst layers;
Safe application intermediate layer: the module that implements all the functionalities of SAI. SAI int mod is relevant for the closed network configuration, while open networks add the SAI cod mod and LFSR (linear shift register, produce pseudo-random numbers series) blocks to the picture;
Safety layer: implements all the functionalities of SL, using CRC module to perform cyclic redundancy check calculations (error detecting codes: cyclic redundancy check, from Table 2).
Access protection layer: implements all the functionalities of APL through the modules AES-CMAC  and AES which provide implementation of the homonymous algorithms.
Adaptation layer management entity: implement all the functionalities of ALE
PVSlib: provides external interfaces to simplify the interactions of the application layer with the protocol.
Figure 5 highlights connections between the layers SAI, SL, and APL with the “ed” lines, realized through shared memory. This notation points out how layers SAI, SL, and APL communicate each other using the interfaces identified as “encapsulation/decapsulation primitive.” More precisely, each of the SAI, SL, and APL layers define (i) an encapsulation primitive, which will be invoked from the upper layer when sending messages and adds the layer-specific header information, and (ii) a decapsulation primitive, which lower layers invoke when receiving messages to verify the correctness of the layer-specific header dataframes. Layers SL and ALE are instead connected by a “TCP” line between their attributes Fake_ALE and Fake_SL. The dashed green rectangle shows how layer ALE could be placed on a remote machine: as a consequence, the communication between ALE and SL is realized through a TCP connection, set up during the initialization of the protocol stack and managed by entities Fake_ALE and Fake_SL.
Information viewpoint elaborates upon the structure and dataflow exchanged between components. To such extent, we specified format, structure, and size of the different PDUs to be exchanged between the PVS layers in the different protocol phases identified by the SL state in Fig. 4. When sending a message through PVS, the User Data crosses the layers SAI, SL, and APL using the encapsulation primitive. After APL encapsulation, the packet is sent by the SL::Fake_ALE and received by the ALE::Fake_SL via TCP through the connection established between the two entities during handshake. When ALE has added its header information, the message can be sent to transport OSI layer.
The system can be seen as an ensemble of objects which exchange data through the interface functions and specification of actions taken by these objects during system’s operative phase. This description includes—but not limits to—generation and verification of safety code, execution cycle and sequence numbers, and more generally sending/receiving algorithms for each layer of the protocol stack. The description of such sending/receiving algorithms causes layer interfaces to be fully defined.
The engineering viewpoint constitutes the most detailed perspective over the system, considering support functions intended to guarantee non-functional properties e.g., security or safety. These aspects are addressed through acceptance/credibility checks (defensive programming: acceptance/credibility checks, from Table 2), control flow monitoring, and cycle redundancy check (CRC).
Acceptance/credibility checks (left part of Fig. 6) have been implemented by checking inputs, intermediate values, and outputs of all the functions through conditional constructs inside each of the SW modules identified during architecture specification.
Control flow monitoring (right part of Fig. 6) includes the software entity monitor, which observes the execution flow with respect to some predefined paths. In a nutshell, the monitoring of the control flow works as follows. The monitor holds primitives, each corresponding to a specific protocol phase (connection establishment, sending and receiving messages, or disconnection) and points to the initial primitive of the current phase. When a layer calls a primitive, it notifies the execution of such primitive to the monitor through the signature transfer function. Then, the monitor verifies the correctness of the evolution of the control flow through the signature checking process, checking if the signature of the primitive transferred by the process conforms to expectations. If either of the checks above fail, they are notified to the application causing connection to close and letting the safety layer entity to return to its initial IDLE (safe) state.
The 64-bit CRC instead ensures integrity of the messages. The choice of CRC with respect to other error detection codes was carried out as CRC is considered proven-in-use for the railway domain [12, 14]. Moreover, as described in [21, 22], the usage of a 64-bit CRC with an adequate generator polynomial (e.g., Jones ) lowers the probability of collisions i.e., same CRC assigned to different dataframes. Polynomial selection is guided by the Hamming distance (HD)  that is defined as the number of bits two dataframes differ. Knowing the minimum HD for specifics dataframes lengths allow us to estimate the probability of undetected errors (Pub) using the following formula :
ϵ : bit error rate of the physical mean (e.g., optic fiber, Wi-Fi);
n : codeword size;
r : code size;
dmin : minimum HD for selected polynomial and dataword size
The formula (1) allows to estimate the probability of undetected errors for a CRC that uses the selected polynomial. In our case, considering as physical mean a plastic optic fiber with 10 GB/s speed and a supposed BER of 10−10 , a maximal dataframes size of 65,000 byte and a minimum HD of 6, we get
Multiplying this value by the maximal supposed throughput of messages and comparing the result obtained against the target tolerable hazard rate per hour (THR), we obtain Pud(ϵ) ≈ 10−27.581 ∗ h−1. Considering that SIL4 requires the failure rate to be less than 10−9, the selected CRC is plenty sufficient under all expected operating conditions, also in case different channels are used and consequently different parameters values are set.
Lastly, the technology viewpoint considers the physical characteristic and assumptions of use (AoU) of the target platform where PVS will be deployed. Our three AoU are mandatory requirements for all the operating systems on which the PVS software will execute: (i) target platform must support multithreading: this is required because our implementation of the protocol requires the execution of multiple thread needed for example to handle the execution cycle or to trigger event-specific handling actions; (ii) target platform must run a POSIX-compliant OS: PVS may execute on multiple platform, and consequently POSIX compliance is a natural choice to improve portability. For this reason, we implemented different layers of the Protocol Stack by using POSIX features like thread and time source management; (iii) target platform must provide TCP/IP networking support: since it is the de-facto standard for communications, we lay PVS on top of the TCP/IP stack.
The concepts and the methodologies defined in the previous section have been detailed during design specification phase, which for each SW module specifies: the scope of the module, the signature of the functions, the interfaces and the data structures. The techniques selected from  for this phase are listed in Table 3.
As pointed out by Table 3, during design specification, we adopted a modular approach in order to implement the architecture’s components depicted in Fig. 7. This technique requires to decompose the software in small parts easy to understand and analyze. Following such design technique, we were able to fully define each software component in terms of interfaces and related parameters. Each of these components has been specified using different modeling techniques, such as the structure diagram of Fig. 7, where all the software components and their relations are identified.
We selected and applied one of the widespread coding standards for safety-critical applications: MISRA C 2012 . Such coding standard specifies a number of guidelines categorized as rules, requirements that are precisely defined and can be enforced solely by analysis of the source code and directives, that instead may not be precisely defined and may require reference to documentation, functional requirements or subjective judgement to check the compliance of the source code.
The compliance of the source code with the MISRA coding standard helps the programmer to prevent undefined behaviors which can arise from an incorrect use of the programming language. Moreover, it ensures the production of a high-quality code which will be easier to understand, to debug and to test.
In general, MISRA C 2012 requires the use of a subset of the standard C99 along with adequate coding styles  (e.g., avoiding libraries such as “stdio.h” to prevent the large number of unspecified, undefined and implementation-defined behaviors associated with streams and file I/O). This contributes to a strongly typed programming language, as MISRA C 2012  requires, among other rules, to explicitly re-define types, e.g., 4, 8, 16-bit integers (Directive 4.6: typedefs that indicate size and signedness should be used in place of the basic numerical types) to be sure of the storage size needed by arithmetic variables on different architectures.
In some cases, the programmer could intentionally violate one or more rules or directive. For example, it could allow the value of a composite expression to be assigned to an object of wider essential type to avoid sub-optimal compiler code generation (violation of Rule 10.6) and obtaining a better performance. In such cases, a deviation must be provided, defining:
The guideline(s) being violated;
The circumstances in which the violation is acceptable;
The reason why the deviation is required;
Background information to explain the context and the language issues;
A set of requirements to include any risk assessment procedures and precautions which must be observed.
In addition to the MISRA coding standard, we selected code metrics and corresponding thresholds that should be checked while developing the C source code of PVS. We choose some of the most acknowledged code metrics in software engineering based on documents [27, 28] and on oral interviews with domain experts. We ended up selecting the metrics lines of code (LoC), cyclomatic complexity (CC), and number of parameters (NoP), while thresholds were extracted from both the experience of practitioners of RFI and on white papers available in the literature . As a result, PVS functions should not exceed 100 uncommented lines (LoC), while the cyclomatic complexity (CC) of each single function should be lower or equal than 10. Finally, the Number of parameters (NoP) should not be greater than 5. Together, such metrics and the respect of their constraints favor a better software quality.
Verification and validation activities
To adhere with CENELEC standards, also verification and validation techniques and activities (shown in Table 4) have been selected and performed in compliance with . It is worth remarking that the software/hardware integration phase is out of scope in our work as it will be privately managed by the owner of the case study. We detail instead each one of techniques applied in Table 4 and in the following subsections that expand on test coverage and static analysis.
Dynamic analysis, testing, and traceability
EN50128  states that activities to be applied for dynamic analysis and testing are: (i) test cases boundary value analysis (to detect the boundary values of the inputs of functions), and (ii) equivalence classes and input partition testing, which selects the correct inputs depending on correlations between input and outputs of each function. Those activities allow deriving unit tests executed as white-box test, i.e., assuming knowledge of the internal structure of functions, and functional tests to be instead executed as black-box tests, i.e., without assuming any knowledge of the internals of functions.
Another white-box testing activity has been the structure-based testing, a testing technique that aims at exercising the largest number of sub-routines in the call graph through an accurate selection of input values. The overall test suite is composed 63 test cases, partitioned as 19 functional test cases, 37 unit test cases and 7 structure-based test cases. Each of these tests include also robustness tests defined by calling functions with unexpected parameters, such as NULL pointers, or numerical overflows like passing 16-bit integers as actual parameters (used during the function call) where the formal parameters (defined in the function signature) were 8-bit integers (see the example test case in Table 5).
For each test case, the approach is the following:
The test case is prepared. In addition to the general data such as the name of the test, requirement and date, the necessary steps to perform the test are identified. The functions called and the relationships between data and expected results are specified.
The data sets to be provided as input to the test are identified.
The test script is written following the test case.
The test is executed. For each data set, the test result is marked as passed or failed based on the correspondence between expected and actual results. When all data sets produce the expected output, the test case is considered passed.
Some tests require a slightly complex procedure: both layers SAI and SL need, to comply with their functional requirements, a pseudo-random number generator. Random number generation has been unit tested as follows: (i) generation of a statistic sample composed by a relevant quantity of random numbers; (ii) use of the ent  library to process the statistic sample, and (iii) evaluation of the statistic measures produced by ent, namely entropy, chi-square mean, arithmetic mean, monte carlo value for Pi, and serial correlation coefficient. The analysis of the random number generation did not reveal any weakness on the target platform, but it is recommended to repeat this test when installing PVS on different platforms.
Another important aspect of dynamic analysis relative to functional testing is given by traceability  that is usually satisfied by filling a traceability matrix, where each functional requirement reported in  is linked to “called functions.” The traceability matrix allows the identification of test cases which needs to be updated in case of change in requirements and to track the overall test execution status.
Finally, the evaluation of test quality should be performed through code coverage, considering two different coverage criteria. The two criteria applied in this project are described below.
Statement coverage: computes the fraction of the executed lines of code with respect to their total. It is a relatively weak criterion since it does not account for logical operators and control structures.
Compound condition coverage: requires that each single condition in a decision should be tested by assigning both the false and the true values
Test coverage reports on the amount of source code that was tested. Coverage depends on specific criteria that can be set depending on the needs of the user and on the characteristic of the system or software. Often, for each selected criteria a certain coverage threshold is set beyond which the test campaign can be considered concluded, that is, the test stop condition. However, the stop condition for our testing activities does not rely on coverage values. Instead, tests should be performed until the exhaustion of the inputs identified through the boundary value analysis and the equivalence classes and input partition testing performed in both dynamic and static analysis. Despite this, it is very useful to analyse coverage, because it gives an indication about how many behaviors have been examined.
Figure 8 reports on test coverage according to the selected criteria, detailing the results for each of the modules depicted in Fig. 5 (apart from the additional sub-module layer_util, added to better modularize the entity layer during the design specification phase).
Coverage is calculated through the tools Gcovr  and Lcov , which profile the code and show code coverage using different criteria, ultimately providing human-readable coverage reports in HTML format.
Noticeably, single coverage values in Fig. 8 rarely reach 100%, with an overall of 89% for statement coverage and 69.8% for compound condition coverage. Such values are determined by the high presence of defensive code: code branches produced by conditional statements needed to check inputs and outputs for acceptance and credibility (acceptance/credibility check) or to verify the evolution of the control flow (control flow monitoring), eventually triggering error handling routines. Given the complexity of designing test capable of cover such branches, we decided to apply manual inspection to verify all the code that was not covered by dynamic analysis.
Static analysis requires the semi-automated analysis of the source code using software tools or manual inspections without executing the code. Compliance with SIL4 requires : (i) control flow analysis, to check the evolution of the control flow, (ii) walkthrough/design reviews, to determine useless or buggy code through manual inspections; (iii) verification of satisfaction of coding rules and metrics, done through the use of the static analysis tool Polyspace . We will report about the control flow analysis and walkthrough performed in the “Control flow analysis and walkthrough” section, reporting also about the result of the evaluation of code metrics and coding rules in “Code metrics” and “Coding rules” sections.
Control flow analysis and walkthrough
Since the monitor object already implements the dynamic control flow monitoring of the whole PVS, we directly step into walkthrough/design reviews (as verification of the control flow is actually a review of the behavior of the Monitor). We asked a colleague, with no knowledge on the project, to act as auditor. He reviewed the flow of execution of the PVS when used in a typical procedure consisting of configuration, connection, sending, receiving and disconnection. This allowed reviewing all the functions implemented (tracked with a checklist). The auditor filled the “Comment” column of Table 6 which address MISRA Directives and Table 7, which points to good practices of software design/development , to eventually report violations of the described rules. Tables 6 and 7 only report rules and practices that were deemed applicable for the walkthrough. As can be seen from the “Comment” columns of Tables 6 and 7, the analysis did not find any violation.
The values of code metrics selected for the implementation of PVS, extracted by using Polyspace , are reported in Table 8. The table reports metric values highlighting in yellow the values that violate the constraints set by the V&V&S plan described in “Design specification” section.
As can be seen from the table above, some of the source files exceed the predefined constraints. The file PVSlib.c violates cyclomatic complexity threshold, while constraints on lines of code are violated in files monitor.c and aes.c. For number of parameters, we have violations in the files PVSlib.c, sai_int_mod.c, sai_cod_mod.c, sai_layer.c, and safety_layer.c. Those violations have been addressed during the finals V&V stages of the project to check whether was possible or necessary to operate some changes on the modules design or implementation. Briefly, the considerations done on each of those violations are:
Cyclomatic complexity: the values could be reduced defining sub-procedures that execute some of the instructions executed by the functions under exam. However, it was not considered necessary to make changes as the two functions were appropriately overhauled and tested; furthermore, their modification would not have brought significant improvements in terms of simplicity or maintainability of the code.
Number of parameters: we did not make any change because during the design specification phase, it was deemed appropriate to keep them unchanged for reasons of logical consistency between the inputs and outputs produced by them.
Lines of code: it has been observed that the tool  calculates the metric as the difference between the brace that determines the start of the function and the brace that determines its end, also counting the empty lines necessary to improve readability of the code. Therefore, the actual metric value for the functions of the monitor.c file was less than the threshold set. The same does not apply to the aes.c file. However, we did not modify its functions to avoid the insertion of bugs and cause malfunctions of the algorithm.
Polyspace  finds violations to MISRA C:2012 coding rules, showing some false positive, that need to be manually dropped, alongside with violations that we do not consider as defects. The final result of running Polyspace on the PVS source code is in Table 9.
After attentive analysis of the violations identified and, on the opportunity to solve them, we opted for motivating deviations. Each deviation describes: (i) the rule; (ii) the use case that triggers the violation; (iii) the root cause for rule violation; (iv) the possible risks given by rule violation; (v) any possible alternative approach, and (vi) a verification code to be used to check risks on different architectures. While aspects (ii) and (v) cannot be shared due to confidentiality, Tables 10 and 11 show deviations respectively for violations of rules 11.3 and 18.8. We do not report about each of the violations identified in Table 9 as deviations are in any case consistent with what shown in Tables 10 and 11.
Performance analysis of the protocol
We conducted a performance analysis targeting throughput in terms of Kilobytes sent and received per second between two remote machine and using three different setups: (i) a normal C99 socket transmission, (ii) PVS closed network, and (iii) PVS open network.
We created bash scripts that exercise PVS by executing the sending and the receiving of 100 packets of different payload sizes, varying from 5 kB to 65 kB, that is the maximum payload allowed by the PVS. The underlying WLAN has a nominal bandwidth of 100 Mbps. Figures 9 and 10 have been generated by averaging the data collected during the execution of the 100 experimental runs for each of the 2 operational modes of PVS and for a basic TCP socket. As expected, the open networks mode of PVS (Category 3 of ) has the lowest throughput. This can be explained considering that this operational mode of PVS embeds AES encryption and decryption, in addition to building or checking a wider packet header. However, the throughput of approximately 300 kB/s is appropriate considering the usual dimension of data exchanged between nodes in the railway network.
In this paper we reported our experience in the implementation of software for safety-critical systems. More in detail, we applied the CENELEC railway standards to deploy the Protocollo Vitale Standard, a safe and secure communication protocol to interlace various railway devices. We described the Protocollo Vitale Standard , examining its structure, and then describing and motivating the techniques used for the design, implementation and V&V processes.
The overall process complies with applicable CENELEC standards [6,7,8], which specify the software lifecycle and the V&V activities to be carried out before certification and decommissioning. While standards EN50126  and EN50159  define the process needed to specify the safety functions to be allocated to the Software, standard EN50128  focuses on technical requirements for the development of software, that needs to adhere with a given safety integrity level (SIL).
For the development of our safety-critical communication protocol, we started from the structured methodology ODP-RM  to define and describe the system architecture, where we applied various modeling formalism like structure diagrams, sequence diagrams, block diagrams, and state machines to describe the system and selected techniques needed to improve the system safety such error detecting codes (CRC), control flow monitoring, and acceptance/credibility checks. In compliance with  and the verification, validation and safety Plan, we implemented our software according to (i) the MISRA C 2012 coding standard  and (ii) a set of software metrics to be respected [27, 28].
The V&V activities consisted then in a thorough revision of each activity done in the previous phases, in the drafting and execution of unit, functional, and structure-based tests, and in the verification of compliance to coding standards and software metrics. Ultimately, we conducted performance analyses to verify the sending/receiving throughput of packets with both open/closed network operational modes of PVS.
As a last remark, we strongly believe that our experience in preparing, implementing, verifying, and validating a software to be installed in safety-critical systems could be used as reference for those who have to deal with safety requirements either for railways or for other domains (e.g., automotive, avionics), which share most of the V&V items with respect to railway standards.
Availability of data and materials
Data is not available due to contractual IPRs.
Avizienis A et al (2004) Basic concepts and taxonomy of dependable and secure computing. IEEE Trans Dependable Secure Comput 1(1):11–33
Ceccarelli A, Zoppi T, Vasenev A, Mori M, Ionita D, Montoya L, & Bondavalli A (2018) Threat analysis in systems-of-systems: an emergence-oriented approach. ACM Transactions on Cyber-Physical Systems 3(2):1–24. USA.
Bhatti ZE, Roop PS, Sinha R (2016) Unified functional safety assessment of industrial automation systems. IEEE Trans Industr Inform 13(1):17–26
Xie G et al (2017) Hardware cost design optimization for functional safety-critical parallel applications on heterogeneous distributed embedded systems. IEEE Trans Industr Inform 14.6:2418–2431
IEC, IEC61508 (2010) 61508 functional safety of electrical/electronic/programmable electronic safety-related systems. International Electrotechnical Commission
CEI EN 50126. Railway applications - the specification and demonstration of Reliability, Availability, Maintainability and Safety (RAMS), 2008.
CENELEC, EN50159 (2010) Railway applications-communication, signalling and processing systems - safety-related communication in transmission systems [Report]: Standard.-[sl]. European Committee for Electro-Technical Standardization
CENELEC, EN50128. Railway applications-communication, signaling and processing systems-software for railway control and protection systems, 2011
CEI EN 50129 (2004) Railway applications - communication, signalling and processing systems - Safety related electronic systems for signalling
Sniady A, Soler J (2012) An overview of GSM-R technology and its shortcomings. In: 2012 12th International Conference on ITS Telecommunications. USA: IEEE; pp 626–629
Fall KR (2011) TCP/IP Illustrated, Volume 1: The Protocols, Addison-Wesley Professional, USA, ISBN: 9780321336316
UNISIG (2015) SUBSET-037 EuroRadio FIS. Version 3.2.0
Winter P et al (2009) Compendium on ERTMS. Eurail Press
SUBSET, UNISIG. 098, RBC-RBC safe communication interface, 2007.
Bertieri D, Zoppi T, Mungiello I, Ceccarelli A, Barbareschi M, Bondavalli A (2019) Implementation, verification and validation of a safe and secure communication protocol for the railway domain. In: LADC 2019 9th American symposium on dependable computing
Wang R, Zhao H-b, Wang S-m (2008) Research on uplink-signal simulator used for BTM test in balise system. J China Railway Soc 30(6):46–50
Technical Committee ISO/IEC JTC 1/SC 7 (1998) ISO/IEC 10746-1:2009 Information technology — Open distributed processing — Reference model: Overview — Part 1
RFI (2017) Relazione Tecnica RFI DTCDNSSS RT IS 05 021 F. Protocollo Vitale Standard. Internal report (in Italian). Italy
PUB, NIST FIPS (2001) 197: advanced encryption standard (AES). Fed Inf Process Stand Publ 197(441):0311
Song, Junhyuk, et al. The aes-cmac algorithm. RFC 4493, 2006.
Wolf J, Michelson A, Levesque A (1982) On the probability of undetected error for linear block codes. IEEE Trans Commun 30(2):317–325
Franekova M, Rastocny K (2011) Modelling of disturbing effects within communication channel for safety-related communication system. Adv Electr Electron Eng 6(2):63–68
Koopman P, Chakravarty T (2004) Cyclic redundancy code (CRC) polynomial selection for embedded networks. In: International conference on dependable systems and networks, 2004. IEEE, pp 145–154
Loquai S et al (2012) 10-Gb/s pulse-amplitude modulated transmission over 1-mm large-core polymer optical fiber. IEEE Photon Technol Lett 24(10):851–853
MISRA C:2012s (2013) Guidelines for the use of the C language in critical systems. MIRA Limited, Warwickshire
Coding Styles. gnu.org/prep/standards/standards.html#Writing-C
Rosenberg LH, Hyatt LE (1997) Software quality metrics for object-oriented environments. Crosstalk J 10(4):1–6
Exida Consulting LLC. C/C++ coding standard recommendations for IEC 61508, version V1, revision R2 (2011)
John Walker, Ent - a pseudorandom number sequence test program. http://fourmilab.ch/random/, 2008.
GCovr Library. https://gcovr.com/en/stable/
LCov Library. https://github.com/linux-test-project/lcov
Polyspace tool, Matlab. https://it.mathworks.com/products/polyspace.html
This work has been funded by RFI Rete Ferroviaria Italiana S.p.A.
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Bertieri, D., Ceccarelli, A., Zoppi, T. et al. Development and validation of a safe communication protocol compliant to railway standards. J Braz Comput Soc 27, 5 (2021). https://doi.org/10.1186/s13173-021-00106-w
- Communication protocol
- Safety integrity level
- Verification and validation