The European Data Format (EDF) is a simple and flexible format for exchange and storage of multichannel biological and physical signals. It was developed by a few European 'medical' engineers who first met at the 1987 international Sleep Congress in Copenhagen. See http://www.edfplus.info/
The original EDF specification has been expanded in several ways. EDF+ supports the addition of annotations and non-continuous recordings. The BioSemi Data Format BDF format uses 24 bits per sample (in stead of the 16 bits per sample in EDF). And BDF+ is an EDF+ like extension of BDF.
This packages supports all these variants.
Both EDF and BDF files consist of an header followed by one or more data records with the recorded signals, either ordinary signals or annotation signals.
This package follows this structure by providing two basic functions: readEdfHeader and readEdfSignals (see the help pages for details)
The examples below are based on the following files:
AFile an EDF+ file with a continuous recording of 2 ordinary signals, two annotation signals and a sub second start specification
BFile a BDF+ file with a continuous recording of 11 ordinary signals and one annotation signal
CFile an EDF+ file with a continuous recording of 11 ordinary signals and one annotation signal
DFile an EDF+ file with a discontinuous recording of 11 ordinary signals and one annotation signal
The BFile, CFile and DFile are derived from the "test_generator_2" test files from http://www.teuniz.net/edf_bdf_testfiles. The AFile is derived from the "test_generator8" file received from Teunis van Beelen by private communications.
libDir <- system.file ("extdata", package="edfReader") AFile <- paste (libDir, '/edfAnnonC.edf', sep='') # a file with 2 annotation signals BFile <- paste (libDir, '/bdfPlusC.bdf' , sep='') # a continuously recorded BDF file CFile <- paste (libDir, '/edfPlusC.edf' , sep='') # a continuously recorded EDF file DFile <- paste (libDir, '/edfPlusD.edf' , sep='') # a discontinuously recorded EDF file
The readEdfHeader function returns a list of class 'ebdfHeader' with all the data from the EDF or BDF file header. Part of this list is a data frame of class 'ebdfSHeader' which contains the signal headers.
A file header can be read with readEdfHeader()
require (edfReader) AHdr <- readEdfHeader (AFile) BHdr <- readEdfHeader (BFile) CHdr <- readEdfHeader (CFile) DHdr <- readEdfHeader (DFile)
Summaries of the header data can be shown with the S3 summary () and print() functions
BHdr
summary (AHdr)
The summary() provides somewhat more information than the print() function. For all details use str().
edfAnnonC.edf contains two annotation. These signals must have the same label. As edfReader names signals after their labels (see below), distinguished names are created by appending '-1' and '-2' to the labels.
NOTE \ \ \ \ \ \ Actually, the startTime fraction .7 is not read from the edf/bdf file header but from the first data record. For details, see the section “Samples, time and periods”.
Summaries of the signal header data can be shown with the S3 print() and summary () functions.
AHdr$sHeader summary (CHdr$sHeader)
The readEdfSignals function with simplify=FALSE returns a list of class 'ebdfSignals' with the signals selected from the EDF / BDF file.
The signals in this list are of the following :
continuous signals
of class 'ebdfCSignal' which may result from all types of EDF(+)/BDF(+) files
fragmented signals
of class 'ebdfFSignal' which result from EDF+D and BDF+D files only
annotation signals
of class 'ebdfASignal’ which may result from all types of EDF+ and BDF+ files.
The signals in an EDF or BDF file can be read with the readEdfSignals function.
ASignals <- readEdfSignals (AHdr) ASignals
NOTE 1\ \ \ \ \ See below in case the file contains only one signal.
NOTE 2\ \ \ \ \ The annotation signal number '1,3' indicates that the 'ebdfASignal' annotation object contains the annotations from both signals 1 and 3. This can be prevented by using the 'mergeASignals=FALSE' option.
DSignals <- readEdfSignals (DHdr) DSignals
By default all fragments in a discontinuously recorded signals will be concatenated into one signal with the gaps filled with NA values. This can be prevented by using the 'fragments=TRUE' option.
DSignalsF <- readEdfSignals (DHdr, fragments = TRUE) DSignalsF
A summary of the list of signals can be shown with the S3 summary () functions. For this 'ebdfSignals' objects it shows the same data as the S3 print function (see above). For object details use the str() function.
The reading of signals can be restricted to specific set by using the signals parameter. Signals can be identified by their signal number, label, name, or signal type ('Ordinary' or 'Annotations'). Also, the list may contain duplicates. The following 3 designations refer e.g. to the same signal.
CSignals8 <- readEdfSignals (CHdr, signals=c(8, "8", "sine 8.5 Hz")) CSignals8
NOTE \ \ \ \ \ \ As in this case only one signal was read, the list of one was simplified to a singe 'ebdfCSignal' object. This could be prevented by using the 'simplify=FALSE' parameter.
EXAMPLE \ \ \ \ \ \ readEdfSignals (CHdr, signals=7, simplify=FALSE)[[1]] and readEdfSignals (CHdr, signals=7) will return the same object.
If required the reading can be restricted also to a particular period. A period is identified with the 'from' and 'till' parameters which specify a time in seconds relative to start of the recording.
ASignalsPeriod <- readEdfSignals (AHdr, from=0.7, till=1.8) ASignalsPeriod
NOTE 1 \ \ \ \ The annotations included are those with an onset up to and including 1.8 sec. The onset for an annotation may be outside the recorded period.
NOTE 2 \ \ \ \ Because the recorded period is only 1.2 sec (see above or use summary(ASignalsPeriod)), the period read will not be [0.7, 1.8) but only [0.7, 1,2).
NOTE 3 \ \ \ \ For time and rounding details see the section “Samples, time and periods”
Summaries of a list of signals can be shown with the S3 print() and summary () functions.
Summaries of a continuously recorded ordinary signal can be shown with the S3 print() and summary () functions.
CSignals <- readEdfSignals (CHdr) summary (CSignals$pulse) # edfReader names signals after their label
Ordinary signals that are not continuously recorded can be read in two different ways:
as a 'continuous' sequence of samples with 'NA' values filling the gaps (use 'fragments=FALSE', the default).
The result is of class 'ebdfCSignal' (the same as for a continuously recorded signal).
as a number fragments of continuously recorded parts (use 'fragments=TRUE')
The result is of class 'ebdfFSignal'.
The latter method will use a more complex data structure, the first may result in a (much too) huge object.
CDSignals <- readEdfSignals (DHdr, from=5.1, till=18) FDSignals <- readEdfSignals (DHdr, fragments=TRUE)
The objects of class 'ebdfCSignal' are printed and summarised in the same way as continuously recorded signals.
Summaries of a not continuously recorded ordinary signal stored in fragments can be shown with the S3 print() and summary () functions.
summary (FDSignals$`sine 8.5 Hz`) # note the "`" quotes for a name with spaces.
NOTE \ \ \ \ \ \ In case of more then 10 fragments, the list of fragments summaries will be limited to 5 followed by the number of omitted summaries.
Summaries of an annotation signal (ASignal) can be shown with the S3 print() and summary () functions.
CSignals$`EDF Annotations` summary(ASignalsPeriod$`EDF Annotations`)
NOTE 1 \ \ \ \ Because both annotation signals are merged into one, the merged signal is named EDF Annotations
again.
NOTE 2 \ \ \ \ The "Record start specs = 0" indicates that the record start specifications are not included, i.e. readEdfSignals was used with the parameter recordStartTimes = FALSE.
In case of an EDF or BDF file, the startTime attribute is based on the startdate and starttime in the EDF/BDF header.
The starttime in an EDF/BDF header is specified up to the second.
In EDF+ and BDT+ files a sub second start time can be specified as the start time of the first data record. See the last paragraph in section .2.2.4 of the EDF+ specification.
See, e.g., the startTime for the
format (AHdr$startTime, format="%Y-%m-%d %H:%M:%OS3", usetz = FALSE) ASignalsPlusStartTimes <- readEdfSignals(AHdr, signals='EDF Annotations-1', recordStarts=TRUE) annots <- ASignalsPlusStartTimes$annotations annots[annots$isRecordStart==TRUE,'onset'][1]
For EDF+ and BDT+ files, the startTime shown by the edfReader is based on both the data in the header and in the start time of the first data record.
As usual, a recording starts at time 0 with sample 1.
Consequently sample n will be at time (n-1)/sRate, where sRate denotes the sample rate.
Apart from rounding errors, a from - till period in readEdfSignals will be the period [from,till), i.e. starting at from and up to be but not including till.
This may sound strange, but this convention has the following properties
the sample rate can be defined as the total number of samples divided by the total time
or, more precise, any period which is a multiple of the sample period sPeriod will always contain the same number of samples
i.e. for any t ≥ 0, [t, t+n*sPeriod) will always contain n samples
for any number of adjacent periods, the total number of samples will be always the sum of the samples in the individual periods.
the first sample will be ceiling(from*sRate) + 1
the last sample will be ceiling(till*sRate)
with from=t and till=t, the empty set [t,t) corresponds to [ceiling(t*sRate) + 1, ceiling(t*sRate)] which is also empty.
In an EDF+/BDF+ file the start of the recording of the signals in a data record is specified in its first annotations signal. For a +D file the recording in a subsequent data record may start at any time after the start of the previous data record plus the record duration time.
The gap between to +D file data records may not be an exact multiple of the sample period of its signals. This raises a question about the alignment of samples in these subsequent data records.
The alignment of ordinary signals can be modelled in two different ways:
A. with a interrupted clock
In which case the first sample for every recorded signal is taken at the start of the recording for that data record
B. which a continuously running clock
In which case the clock starts at the start of the first record and all sampling is based on the individual signal sample rate(s). In other words,
the sample time for a signal sample in the data records is aligned to its sample rate.
It should be noted that the model with the continuously running clock is (implicitly) required if one want to map all fragments of a signal into a single single (with NA values in the gap) one.
The edfReader supports both models, for any fragment it provides both the aligned data according to the "continuously running clock" model as well as the record start time from the data record.
The first sample in every +D record can be aligned as follows:
n = ceiling (sRate * recordStartTime) + 1
and the
sampleTime = (n - 1) / sRate
Where
sRate is the sample rate
n is the sample number when continuously sampled from the start of the recording
recordStartTime denotes the start of the record relative to the start of the recording
sampleTime the (aligned) sample time for the n's sample
However, in order to avoid rounding unnecessary rounding errors the first formula is actually implemented as follows:
n = ceiling (sRate * (recordStartTime - maxTErr)) + 1
Where maxTErr = 5 * .Machine$double.eps (= normally 5 * 2.220446e-16 = 1.110223e-15 )
For any signal object and any fragment:
the 'start' parameter contains the aligned signal / fragment start time in seconds, relative to the start in the recording (for a signal it is always 0)
the 'fromSample' parameter contains the sample number for the corresponding sample
the 'recordingStart' parameter contains the start of the recording for the data record. For a '+' file this is the time specified in the first annotation signal
a 'from' parameter in an readEdfSignals invocation is relative to the start of the recording. The default value is 0
a 'till' parameter in an readEdfSignals invocation is relative to the start of the recording. The default value is Inf
a 'recordedPeriod' parameter equals the number of data records times the record duration time as specified in the header
a totalPeriod parameter contains the period from the start time in the header till the end of the recording, i.e. the record start time of the last data record plus the record duration time
The header data encompass the following:
str (CHdr, max.level=1)
The fields version, patient, recordingId, startTime, headerLength, reserved, nRecords, recordDuration, and nSignals are from the file header.
the startSecondFraction attribute contains the sub second start data specified in the first data record. See the 'The start time' section above.
NOTE \ \ \ \ \ \ In order to avoid annoying rounding errors, the sub second start time is stored separately. trunc (startTime, 'sec') is less accurate.
sHeader is detailed below. The others attributes are derived ones.
The signal header data encompass the following:
str (CHdr$sHeader, max.level=1)
The fields label, transducerType, physicalDim, physicalMin, physicalMax, digitalMin, digitalMax, preFilter, samplesPerRecord, and reserved are from the file header. The others are derived. Gain and offset are used to map the digital sample values to the range of physical values.
For annotation signals the only relevant fields are "label" which must have the value "EDF Annotations" (or "BDF Annotations") and "samplesPerRecord".
The data for ordinary signal objects of class 'ebdfCSignal' encompass the following:
str(CSignals$pulse, max.level=1)
The attributes startTime, signalNumber, label, isContinuous, isAnnotation, recordedPeriod, totalPeriod, transducerType, sampleBits, sRate, range, and preFilter are (derived) from the header data.
For a continuously recorded signal the totalPeriod is equal to the recordedPeriod. For a not continuously recorded signal the total period equals the start of the last data record plus its duration.
The attributes from and till contain the values of the corresponding actual readEdfSignals parameters. The default values are 0 and Inf.
The start attribute contains the start time and is always zero. The fromSample attribute contains the first sample number and is always 1. By including these attributes, signals and fragments share the same time/sample attributes.
The signal attribute contains the sample data from the EDF / BDF data records. If read with the readEdfSignals parameter physical=TRUE, the default, the digital sample values are mapped to physical values. With physical=FALSE, signals contains the digital sample values.
The physical values are calculated as follows:
physicalValue = gain * digitalValue + offset,
with:
gain = (physicalMax - physicalMin) / (digitalMax - digitalMin)
offset = physicalMax - gain * digitalMax
The data for ordinary signal objects of class 'ebdfFSignal' encompass the following:
str(FDSignals$`sine 8.5 Hz`, max.level=1)
For the attributes startTime, signalNumber, label, isContinuous, isAnnotation, recordedPeriod, totalPeriod, from, till, start, fromSample, transducerType, sampleBits, sRate, range and preFilter see the previous section.
For a not continuously recorded signal the total period equals the start of the last data record plus its duration.
The fragments attribute contains the list of recorded fragments.
The data of a signal fragment in objects of class 'ebdfFSignal' encompass the following:
str(FDSignals$`sine 8.5 Hz`$fragments[[1]], max.level=1)
The fromSample attribute contains the sample number of the first sample in this fragment (as if it were a continuous recording). The start attribute the sample time for this sample.
The 'signal' attribute contains the fragment's sample values. These may be physical values (the default) or digital values (see above).
The data for annotation signal objects of class 'ebdfASignal' encompass the following:
str(ASignals$`EDF Annotations`, max.level=1)
For the attributes startTime, signalNumber, label, isContinuous, isAnnotation, totalPeriod, from, and till see the section for objects of class 'ebdfCSignal'.
The annotations attribute contains a data frame with the individual annotations.
The data for a single annotation encompass the following:
str(ASignals$`EDF Annotations`$annotations, max.level=1)
The record attribute refers to the data record the annotation was read from.
The onset attribute contains the time of the annotation relative to the start of the recording.
The duration attribute contains the duration of the annotated event.
The isRecordStart indicates whether or not this annotation is the first one in a data record (and indicates the start time of the recording of that record).
The annotation attribute contains the annotations associated with the onset and duration.
The fromSignal attribute, if present, refers to the signal that contains the annotation. This attributed is present only if the annotations from different signals were merged into one ‘ebdfASignal’ object.
One of the the first things you may want to do with the imported signals is to have look at them.
This can be acheived with one of the several plot packages available, e.g.:
- ‘plot’ (included in the base installation) - ‘lattice', or - 'ggplot2'
As an example, the function plotEdfSignals() uses ggplot2 and will plot one or more signals
over some period of time. The plot shown presents the period from 0.2 till 0.5 seconds of the
'sine 8.5 Hz' and the 'sine 50 Hz' signals in CSignals.
plotEdfSignals <- function (signals,labels, from=0, till=Inf) { nLabels <- length (labels) sRate <- numeric (length = nLabels) fromS <- integer (length = nLabels) tillS <- integer (length = nLabels) sLength <- integer (length = nLabels) for (i in 1:nLabels) { sRate[i] <- signals[[labels[i]]]$sRate fromS[i] <- ceiling (sRate[i] * max (from, 0)) +1 tillS[i] <- ceiling (sRate[i] * till) tillS[i] <- min (tillS[i], length(signals[[labels[i]]]$signal)) sLength[i] <- tillS[i] - fromS[i] + 1 } totLength <- sum (sLength) cat (" totLength=", totLength) time <- numeric (length = totLength) signal <- numeric (length = totLength) label <- character (length = totLength) from <- 1 for (i in 1:nLabels) { till <- from + sLength[i] - 1 time [from:till] <- seq (from=fromS[i]-1, to=(tillS[i]-1)) / sRate[i] signal[from:till] <- signals[[labels[i]]]$signal[fromS[i]:tillS[i]] label [from:till] <- rep(labels[i], sLength[i]) from <- till + 1 } cat (" | from-1=", from-1,'\n') ggplotDF <- data.frame (time=time, signal=signal, label=label) ggplot (ggplotDF, aes(x=time, y=signal, colour=label)) + geom_line() } if (require(ggplot2)) { CSignals <- readEdfSignals (CHdr) plotEdfSignals (CSignals, labels=c('sine 8.5 Hz', 'sine 50 Hz'), from=.2, till=0.5) }
For details about ggplot2 see e.g. the 'R Graphics cookbook'
Enjoy.
This package has used code from
edf.R version 0.3 (27-11-2013) from from Fabien Feschet,
http://data-auvergne.fr/cloud/index.php/s/WYmFEDZylFWJzNs
the work of Henelius Andreas as of July 2015,
https://github.com/bwrc/edf
Specification of EDF
http://www.edfplus.info/specs/edf.html
Specification of EDF+
http://www.edfplus.info/specs/edfplus.html
Specification of EDF++
http://195.154.67.227/en/contribute/edf/
Specification of BDF
see 'Which file format does BioSemi use' at
http://www.biosemi.com/faq/file_format.htm
Specification of BDF+
http://www.teuniz.net/edfbrowser/bdfplus%20format%20description.html
Other useful EDF related sources can be found at:
http://www.edfplus.info/downloads/
/
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.