1 Recommended Practice for Signal Treatment Applied to Smart Transducers ISO/IEC/IEEE Sponsoring Society and Committee: IEEE Industrial Electronics Society/Industrial Electronics Society Standards Committee (IES/IES) Sponsor Chair: Victor Huang Joint Sponsor: IEEE Instrumentation and Measurement Society/TC9 - Sensor Technology (IM/ST) Chair Kang Lee- NIST-USA Working Group Chair – IES: Gustavo Monte (Universidad Tecnologica Nacional, Argentina) Working Group Co-Chair – IMS: Ruqiang Yan (SouthEast Univ, China) Recommended Practice for Signal Treatment Applied to Smart Transducers

2 ISO/IEC/IEEE Development Status The purpose is to define a standardized and universal framework that allows smart transducers to extract features of the signal being generated and measured. With the definition of these practices, the raw data can be converted into information and then into knowledge. In this context, knowledge means understanding of the nature of the transducer signal. This understanding can be shared with the system and other transducers in order to form a platform for sensory knowledge fusion. PURPOSE

3 Recommended Practice for Signal Treatment Applied to Smart Transducers ISO/IEC/IEEE SCOPE Scope: This recommended practice defines signal processing algorithms and data structure in order to share and to infer signal and state information of an instrumentation or control system. These algorithms are based on their own signal and also on the transducers attached to the system. The recommended practice also defines the commands and replies for requesting information and algorithms for shape analysis such as exponential, sinusoidal, impulsive noise, noise, and tendency.

4 Recommended Practice for Signal Treatment Applied to Smart Transducers The future of smart sensors and Why we need a standard for sensor signals Smart Sensor Data transmission Wired or Wireless Bottleneck (channel capacity, regulations,spectrum,power constrains) More processing power.For example, The Play Station 3 has the same peak processing (1.8 Teraflops) than the Sandia Lab Supercomputer in (IEEE SPECTRUM, February 2011 pag, 48) In the near future, the sensor signal will be entirely processed into the sensor, even for small and cheap sensors. Today, This is done but only for complex signal processing related to specific issues. In general, we are wasting processing power. POM Point Of Measurement (The world is inferred from here) POA Point Of Adquisition

5 Recommended Practice for Signal Treatment Applied to Smart Transducers ISO/IEC/IEEE Development Status Assuming that the sensor signal will be processed in the POM. What the sensor should report? DATA INFORMATION KNOWLEDGE BANDWIDTH Data=> Samples Information=> Preprocessed data without a particular objective. Knowledge-=> Feature extraction of particular interest. ( the lowest bit rate is obtained with questions with yes/no answer that require signal processing) This standard could be very important for the IoT

6 Recommended Practice for Signal Treatment Applied to Smart Transducers Actuator(Heater) Temperature Sensor Actuator(Heater) Does your signal exhibit an exponential behavior? yes You are validated Actuator-Sensor dialogue Brand ABrand B Thank You! Working properly state The whole system is validated, actuator OK, temp sensor OK, liquid inside If something is wrong, we realize of it before a timeout!. Few bits/sec EXAMPLE

7 P Status PAR Request Date: 18-Oct-2011 PAR Approval Date: 06-Feb-2012 PAR Expiration Date: 31-Dec WG P&P Working Group Policies and Procedures ( based on IEEE templates) 2.Conform a main WG ( In this case from IES and IMS) 3.Initial Draft document – (Important to establish the ideas that fired the proposal) 4.Call for Participation (IEEE SA) 5.Meeting (WEB based).(aprox every two months). Occasionally face to face meetings. IES and I&MS are co-sponsors. IMPORTANT DATES STEPS: Recommended Practice for Signal Treatment Applied to Smart Transducers

8 OUTINE OF THE PROPOSED STANDARD The smart sensor should recognize the evolution of its signal like an observer does. The sensor should see the “big picture” Recommended Practice for Signal Treatment Applied to Smart Transducers

9 OUTINE OF THE PROPOSED STANDARD RTSAL Algorithm The proposal is to standardized the sensor signal behavior based on time domain analysis for any kind of sensor. Recommended Practice for Signal Treatment Applied to Smart Transducers

10 OUTINE OF THE PROPOSED STANDARD Establish the relationship between tagged samples. This can be seen as a new concept for sampling process. Recommended Practice for Signal Treatment Applied to Smart Transducers

11 OUTLINE OF THE PROPOSED STANDARD Three vectors describe the signal: (M,C,T) M (Mark)={ 1.8,1.7,-0.3….. C (Class)={g,g,e,d,d,g,e…. T (TempPos)={6,8,12,14,……… Recommended Practice for Signal Treatment Applied to Smart Transducers

12 DIAGRAM OF THE PROPOSED STANDARD Recommended Practice for Signal Treatment Applied to Smart Transducers

13 OUTLINE OF THE PROPOSED STANDARD: EXPONENTIAL DETECTION INPUT PARAMETERS: Number of consecutive segments required to start the algorithm. Number of samples for predicted signal. OUTPUT PARAMETERS: Exponential detection (yes/no). Type of detection. ( four types of exponential shape) Steady state value. Predicted signal (yes/no), value. The steady state value can be predicted Recommended Practice for Signal Treatment Applied to Smart Transducers

14 OUTLINE OF PROPOSED STANDARD: NOISE DETECTION Noise detection Noise is detected by computing the distance in samples among maximums and minimums. If a maximum occurs, then the next one is a minimum and vice versa. If the temporal difference between these two singularities is less than a prefixed value for N times, a noisy flag is turned on. INPUT PARAMETERS: Maximum amount of samples between max-min-max to be considered as a noisy signal. Number of consecutive detection of noisy signal. OUTPUT PARAMETERS: NOISY (yes/no) Local minimum at the union of: fd, fg,ed, eg segments. Local maximum at the union of df,de,ge,gf. Recommended Practice for Signal Treatment Applied to Smart Transducers

15 OUTLINE OF PROPOSED STANDARD: IMPULSIVE NOISE DETECTION An abrupt reduction of the segment length starts the algorithm. This indicates the presence of noise. To be considered as impulsive noise, the change in amplitude must be significant. Once the algorithm has started, it looks for the first maximum or minimum. Then, it computes the amplitude change from the value before the length reduction. The length of the impulsive noise is computed from the temporal difference between the first segment and the last one. For example, the sequence “dddffff” indicates the occurrence of a maximum, and the impulse length is the difference between the temporal position of the last “f” and the first “d”. INPUT PARAMETERS: Relative amplitude change to be considered as an impulsive noise. OUTPUT PARAMETERS: Detection yes/no. Type of impulse (“df”,”eg”….) Amplitude of impulse. Length of impulse in samples. Recommended Practice for Signal Treatment Applied to Smart Transducers

16 OUTLINE OF PROPOSED STANDARD: SINUSOIDAL PATTERN DETECTION Dots indicate tagged samples. Class(n)=[ g,g,g,e,e,e,f,f,f,f,g,g]. INPUT PARAMETERS: None OUTPUT PARAMETERS: Detection of pattern (yes/no) Estimated Period (samples) Stable or unstable oscillations. The sinusoidal patter is composed of the four class segment “gefd” including repetitions of the same class. For example “gggeeefffdddd” is also a valid sinusoidal pattern. The maximum occurs at the union of “ge” segments and the minimum at “fd”. The period is the difference between singularities of the same kind. By computing peak to peak values, we can determine if the oscillations are increasing or decreasing. Recommended Practice for Signal Treatment Applied to Smart Transducers

17 OUTLINE OF PROPOSED STANDARD: TENDENCY ESTIMATION INPUT PARAMETERS: Windows size in samples. OUTPUT PARAMETERS: Tendency for max.[-1,1] Tendency for min.[-1,1] Tendency is computed taking into account the increment or decrement of both, the maximums and minimums in a temporal window. Growing trend is normalized to +1 when all the new maximums are higher than the previous ones in the window. On the other hand, the trend is decreasing, and normalized to -1, when all the new maximums are lower than the previous ones in the window. The same procedure is executed for minimums. Recommended Practice for Signal Treatment Applied to Smart Transducers

18 Recommended Practice for Signal Treatment Applied to Smart Transducers OUTLINE OF PROPOSED STANDARD: MORE RELIABLE VALUE Since that there is information about the signal shape, it is possible to exclude in the mean estimation of the signal some artifacts. In this algorithm, segments that belong to impulsive noise are discarded in the mean computation. INPUT PARAMETERS: Window length for mean computation. OUTPUT PARAMETERS: Mean including all the segments. Mean without impulsive noise. Low pass filtered signal

19 Recommended Practice for Signal Treatment Applied to Smart Transducers OUTLINE OF THE PROPOSED STANDARD: COMPLEX PATTERN DETECTION Detection of normal QRS complex in ECG signals Unexpected sequence of segments in a real signal. In this case too many “f” and “d” types.

20 Recommended Practice for Signal Treatment Applied to Smart Transducers COMPLEX PATTERN DETECTION: EEG spike detection with noise EEG signal (eeg_3_01) from MatLabEEG signal with normal noise added EEg after 3 iterations of RTSAL err=0.001EEg after 10 iterations of RTSAL err=0.001

21 Recommended Practice for Signal Treatment Applied to Smart Transducers OUTLINE OF PROPOSED STANDARD: filtering Sinusoid detection with noise. FS=22050, f=345 Hz SNR=7dB. Iterations=10. Interpolation error: 1%. Red: original signal. Green: signal reconstructed from MCT vectors. A new set M’C’T’ is obtained.

22 Recommended Practice for Signal Treatment Applied to Smart Transducers Working Group objectives We will try to keep the smart sampling process (MCT vectors) as a platform for every algorithm in this WG. Review and define the proposed algorithms for layer one. Propose new algorithms. Develop user defined application code based on MCT vectors. Propose the data structure including commands scheme. Propose a time synchronization scheme. Define and propose the interface with the IEEE 1451 standards. Propose learning algorithms for specific patterns. Propose filtering and signal compression.

23 Recommended Practice for Signal Treatment Applied to Smart Transducers Working group organization Coordination subgroup IES &IMS Impulse, noise detection mean estimation. Tendency, exponential and sinusoidal patterns. Interface with IEEE 1451 Standards and Data structure Filtering, signal compression and prediction New and learning Algorithms. User Defined Application Code Testing Subgroups

24 Recommended Practice for Signal Treatment Applied to Smart Transducers Working group organization Subgroups SUBGROUP SG1 Objectives: To define, propose, and validate first layer algorithms based on MCT vectors. Review the initial proposal, modify it and suggest changes. LEADER: Dr. Antonio Pietrosanto - University of Salerno- ITALY SUBGROUP SG2 Objectives: To define and coordinate commands and data structure consistently with IEEE 1451 standards. Review and propose a time synchronization scheme for data. Normalize name of variables and data. LEADER: Dr. Eugene Song- National Institute of Standard Technology (NIST) - USA

25 Recommended Practice for Signal Treatment Applied to Smart Transducers Working group organization SUBGROUP SG3 Objectives: To define, review and propose algorithms based on MCT vectors and first layer outputs for filtering, signal compression and prediction. LEADER: Dr. Ruqiang Yang - School of Instrument Science and Engineering, Southeast University, Nanjing, China SUBGROUP SG4 Objectives: To propose new algorithms that the group considers that they must be included. Define the data structure for embedding user defined application code and pattern learning algorithms. LEADER: Dr. Zheng Liu- Toyota Technological Institute- JAPAN SUBGROUP SG5 Objectives: To test and validate proposed algorithms using simulation software (MatLab, Octave, Scilab…) and C code for microcontrollers. LEADER: Dr. Vincenzo Paciello-Università degli Studi di Salerno- ITALY

26 Recommended Practice for Signal Treatment Applied to Smart Transducers Thank you!