RF Testing of WLAN Products

Application Note 1380-1

Wireless Local Area Networksallow mobile computer users toremain connected to a network,and access resources, while on themove or physically disconnectedfrom the network. Over the pastseveral years, different WirelessLocal Area Network (WLAN)technologies and standardshave been developed. Two ofthe major driving forces behindthese standards are the Instituteof Electrical and ElectronicEngineers (IEEE) and the EuropeanTelecommunications StandardsInstitute (ETSI). From theseorganizations have emerged twoof the most successful WLANstandards, IEEE 802.11 and ETSIHIPERLAN. Recent advances intechnology have enabled theproduction of affordable andreliable networking hardwarefor use in wireless LANs.

The acronyms used in thisdocument are either defined attheir first usage or in the glossaryon page 38.

This application note looks at themodulation technology behindseveral WLAN standards and themeasurement techniques that canbe used to troubleshoot andquantify their RF performance.The emphasis will be on 802.11b,802.11a, and HIPERLAN Type 1and Type 2. The principal focus ofthis document is the physical RFlayer of WLAN signals, as opposedto the MAC layer or higher layersof a WLAN signal. This includestime-, frequency-, and modulation-domain analysis and trouble-shooting, as well as the basicmodulation theory behind thesestandards.

Various modulation schemes areimplemented in the standards,and this document includesinformation on FSK, MSK, GMSK,CCK, and OFDM modulation.

For more information on making IEEE 802.11n MIMOmeasurements, see ApplicationNote 1509, Agilent MIMO WirelessLAN PHY Layer (RF) Operationand Measurement, publicationnumber 5989-3443EN. Forinformation on WiMAXmeasurements, see ApplicationNote 1578, IEEE 802.16e OFDMASignal Measurements andTroubleshooting, publicationnumber 5989-2382EN.

Introduction

2

1. HIPERLAN Type 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1 FSK modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 GMSK modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 GMSK demodulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2. 802.11 and 802.11b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1 802.11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 802.11b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.3 Modulation analysis of 802.11b signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3. 802.11a and HIPERLAN Type 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.1 OFDM signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.2 Dealing with multipath . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.3 Modulation analysis of an OFDM signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

Acronym Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

WLAN Standards Summary Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Table of Contents

1.1 FSK modulationHIPERLAN Type 1 uses twodifferent modulation formats: asimple 2-level FSK format for low-rate data, and 0.3 GMSK for high-rate data. Since 2-level FSK is thesimplest modulation formatdiscussed in this application note,it is an appropriate place to start.

The 2-level FSK format uses asymbol rate of 1.4705875 Mbit/s.In this case this is also the bit rate.The data is used to produce aRECT-filtered signal, which maybe passed through another low-pass filter to limit the signalspectrum. The low-pass filteredsignal is then fed to an FMmodulator with a frequencydeviation of ±368 kHz. Thetransmitter filters must be

carefully designed becausenarrow filters will help containthe spectrum, but they will alsoslow down the frequencytransitions between symbols.

Figure 1.1.1 shows a screen plotfrom the Agilent Technologies89600 vector signal analysis (VSA)software with the characteristicsof the 2-FSK signal. The upper leftgrid is the FM demodulated signal.The lower left grid shows thespectrum of this particularsignal, which was generated byan Agilent Technologies ESGsignal generator. The instrumentspan is set to 18 MHz. The choiceof span involves trading off the noise bandwidth of themeasurement with the speed ofthe symbol transitions. Using a

wider span introduces morenoise into the measurement andnarrower spans eliminate higherfrequency components in thesignal and increase the durationof the FSK transitions. Thespecification calls for a 50 nsectransition, as will be shown later.To measure this transition, thespan should not be much less than1/50 nsec, or 20 MHz. The upperright diagram shows the eyediagram. The markers are spacedabout 68 nsec apart. The lowerright trace contains summaryinformation and is one of the mostuseful traces. It contains the rawbits, the deviation, the FSK (FM)error, and the carrier frequencyoffset. The FSK error is an RMSmeasure of the residual FM errornormalized by the deviation.

3

1. HIPERLAN Type 1

Figure 1.1.1. 2-FSK signal characteristics

Figure 1.1.2 shows a typical eye-diagram mask. The eye-diagrammask serves two purposes. First,it places limits on the deviationand, second, it places limits on thespeed of the symbol transition.Taking the signal two symbols ata time and overlaying them oneon top of the other creates aneye diagram.

Important FM modulatorcharacteristics that affect the eye-diagram are deviation,modulator bandwidth, filtercharacteristics, and frequencystability. If the modulatorbandwidth is insufficient, therise/fall times will be too slow.The baseband and IF filters canalso cause this problem as wellas overshoot and ringing.

4

Figure 1.1.2. Eye-diagram mask for low-rate FSK modulation

Residual FM is undesired FMmodulation caused by spurioussignals getting into the modulator.For example, in Figure 1.1.3,the Agilent E4438C ESG’s FMmodulator was used to add a20 kHz (rate and deviation) FMsignal to the FSK waveform.The sinusoidal error is shown inthe lower trace. The error is largeenough that you can actually seethe entire FSK signal moving upand down.

Some FM modulators have abaseline drift problem. Thiscan occur, for example, whenseveral identical bits are sent ina row (e.g. 3 ones). The carrierfrequency drifts in a positivedirection for a sequence of onesand in a negative direction fora sequence of zeros. It is easilydetected by looking at the FMsignal plot with the FSK errorplot, as configured above.

5

Figure 1.1.3. Residual FM on a FSK waveform

1.2 GMSK modulationFor high-rate data, HIPERLANType 1 uses a 23.5294 Mbit/sGMSK signal. There are twoways to generate the GMSK signal.The first method uses an FMmodulator, as shown in Figure1.2.1. A mistake that designersoften make when working withbaseband signals is exciting thebaseband filter with RECT pulsesinstead of Dirac deltas. The useof RECT pulses is equivalent toadding a second filter, whoseresponse is a rectangular impulse,T seconds wide. This causes the baseband spectrum to have an extra sin(x)/x roll off.

While simple to implement, theuse of an FM modulator for GMSKis usually not a good idea as itrequires the use of coherentdemodulators. Coherentdemodulators require excellentphase control. Recognizing this,the ETSI modulation qualitymeasurement for this signal is ameasure of the RMS phase error,just as it is for GSM. The problemwith an FM modulator iscontrolling the frequencydeviation. Ignoring ISI, thefrequency deviation needs to beexactly 1/4th of the symbol rate toproduce a 90 degree phase rotationin one symbol interval.

In systems where non-coherentreceivers are used, this modulationis referred to as GFSK (as in 802.112- and 4-level GFSK). For GFSK,the deviation is not strictly limited to 1/4th of the symbol rate. It isworth noting that the 0.3 Gaussianfilter is not zero-ISI, as can beseen by the frequency trajectoryfor 1-0-1 or 0-1-0 transitions. Onlywhen two identical bits are sent in a row does the signal reach full deviation.

6

Figure 1.2.1. GMSK modulation method 1

7

The second approach to GMSKmodulation is to run the signalrepresenting frequency into anintegrator, shown in Figure 1.2.2,which creates a signal thatrepresents phase and thenpassed to a phase modulator –usually implemented using an I/Qmodulator. Normally, everythingup to the I/Q modulator will bedone in DSP. Relative to the FMmodulator approach, this methodfor generating GMSK signals haswell-controlled phase.

There is little to go wrong withthis approach prior to the I/Qmodulator, with the possibleexception of integration errors(and the inadvertent use of aRECT filter), because the signalhas a constant envelope. Themajority of the analog problemsare with the I/Q modulator or thelocal oscillators. Errors that maycontribute to RMS phase errorinclude LO feed through, gain andphase imbalance, LO phase noise,and spurious signals.

Figure 1.2.2. GMSK modulation method 2

1.3 GMSK demodulationFigure 1.3.1 shows a plot of theAgilent 89640A VSA measuringa 23.5294 Mbit/s GMSK signal.This signal was intentionallygenerated with 5 degrees ofquadrature skew. The Agilent89640A VSA has 36 MHz ofinformation bandwidth, which issufficient to measure the signal.The main lobe of the spectrum,about 33 MHz wide, is visible inthe lower left grid.

The upper left grid shows theconstellation. The oval shape ofthe polar plot clearly shows thequadrature skew. The dots on the

perimeter of the oval representthe symbol points and are spacedevery 90 degrees, as expected.A closer inspection reveals thatthere are actually 4 clusters of 3symbol dots. The 3-dot areas are aresult of the ISI introduced by theGaussian filter.

The upper right grid is a plotof the phase error, shown over200 symbols. The waveformrepresents the difference betweenthe measured carrier phasetrajectory and the ideal phasetrajectory, computed using thedetected bits. The RMS of thisphase-error signal is the

modulation quality metric.The lower right grid shows theRMS phase error as approximately2 degrees (the allowable limit is10 degrees) and a peak phaseerror of 3 degrees (the allowablelimit is 30 degrees). The table alsoreports that there are 4.6 degreesof quadrature skew and 0.25 dBof gain imbalance. While there areno limits set on skew and imbalancein the standard, it is useful toknow what these numbers are,since these errors contribute tothe phase error metric.

8

Figure 1.3.1. GMSK signal demodulation

The Agilent 89640A VSA includes2- and 4-level FSK demodulators.These are useful for GFSKwaveforms such as those foundin 802.11. The GFSK demodulatorcan also be used on GMSKwaveforms, like the high-ratesignal in HIPERLAN Type 1.Errors such as carrier frequencyinstability, or settling, are easierto see as an FM error than theyare as a phase error. The FM (orFSK) error is the differencebetween the measured frequencytrajectory and the ideal frequencytrajectory based on thedemodulated bits.

Figure 1.3.2 shows (top to bottom,left to right) a display of the FSKeye diagram, the FM waveformover 20 symbols, the FM errorover those same 20 symbols, andthe summary table with informationsuch as carrier frequency offsetand deviation. As expected, thedeviation is 1/4th of the symbolrate. Notice how the ISI created bythe Gaussian filter produces sixlevels in the eye diagram. Thissignal is obviously more heavilyfiltered than the low-rate FSKsignal.

9

Figure 1.3.2. GMSK signal demodulated as a GFSK signal

2.1 802.11The original 802.11 standardsupports data rates at 1 Mbit/secand 2 Mbit/sec. This is donethrough 2- and 4-level GFSKfrequency hopping spreadspectrum (FHSS), or throughBPSK and QPSK direct sequencespread spectrum (DSSS). Thissection will focus on DSSS.

In spread-spectrumcommunications, the signalbandwidth is increased (in thiscase of DSSS, by a factor of 11)

without increasing the data rate.This is done to increase immunitytoward interference from hostilemicrowave ovens, multipath, andco-channel signals. Spreadspectrum technology allows thechip rate to remain constant whilethe data rate changes to matchconditions. For example, the1Mbit/sec rate has a spreadingfactor of 11, using BPSKmodulation. QPSK replaces BPSKin order to double the bit rate to2 Mbit/sec, at the same spreadingrate of 11.

As shown in Figure 2.1.1, to createa DSSS signal, a lower rate signalis multiplied by a higher ratesignal. For 1 Mbit/sec, the 1 MHz(D)BPSK signal is multiplied by an11 MHz BPSK signal. Although not generally true of DSSS signals,for this particular signal it can besaid that the input bits determinethe phase rotation of the spreadingcode or that it is a (D)BPSK datasignal spread by a BPSK spreadingsequence, producing a BPSKconstellation.

10

2. 802.11 & 802.11b

Figure 2.1.1. DSSS signal spreading for 1 Mbit/sec data rate

BPSK

While any spreading sequencecould be used, sequences areusually chosen for their spectralproperties and for low crosscorrelation with other sequenceslikely to interfere. For 1 and2 Mbit/sec 802.11, an 11-chipBarker sequence, is used. Thevalue of the autocorrelationfunction for the Barker sequenceis 1, -1, or 0 at all offsets exceptzero, where it is 11. This makesfor a more uniform spectrum,and better performance in thereceivers.

To double the bit rate from1 Mbit/sec to 2 Mbit/sec, the bitsare taken two at a time. The phaserotation now has four rotations: 0,90, 180 and 270 degrees, insteadof being either 0 or 180 degrees.The symbol rate remains at1 Msym/sec. This can be thoughtof as a (D)QPSK signal spread by a BPSK signal to produce aQPSK signal.

As mentioned previously, at1 Mbit/sec, an 11-chip Barkersequence with two-phaserotations is used to produce onesymbol with two-phase states.At 2 Mbit/sec, the same sequenceis used with four phase rotationsin order to produce one symbolwith four phase states. If this

approach were extended, 2048rotations of an 11 bit sequencewould be needed for a data rateof 11 Mbit/sec specified in802.11b. Obviously, a differentapproach is needed.

While it would be possible to usesome form of a QAM constellationinstead of PSK, this would havethe undesirable side effect ofincreasing the peak-to-averageratio of the signal, makingamplification more difficult.The solution, used in 802.11b, isto retain a QPSK constellation atthe transmitter output. The resultis CCK, or Complementary CodeKeying. The complementary codesare a set of nearly orthogonalcomplex sequences.

11

Figure 2.1.3. Complications arise when trying to achieve an 11 Mbit/sec data rate

Figure 2.1.2. DSSS signal spreading for 2 Mbit/sec data rate

2.2 802.11b802.11b adds 5.5 Mbit/sec and11 Mbit/sec data rates to the 802.11standard. In a rather odd twist,the progression of technology inthe 802.11 standard is 802.11 to802.11b to 802.11a, and eventuallyto 802.11g. Many find the fact that802.11b is slower than 802.11aconfusing.

802.11b, like 802.11 described inthe previous section, is designedfor use in the 2.4 GHz ISM band.The technology behind 802.11b isdirect sequence spread spectrumusing complementary code keying(CCK). Changing both the spreadingfactor and/or the modulationformat varies the bit rate.

To achieve 5.5 and 11 Mbit/secrates, the spreading length isfirst reduced from 11 to 8. Thisincreases the symbol rate from1 Msym/sec to 1.375 Msym/sec.To achieve 5.5 Mbit/sec bit ratesusing the 1.375 MHz symbol rate,one needs to transmit 5.5/1.375or 4 bits/symbol. For 11 Mbit/secone obviously needs 8 bits/symbol.

The approach taken for 802.11b,which keeps the QPSK spreadspectrum signal and still providesthe required number of bits/symbol,uses all but two of the bits toselect from a set of spreadingsequences. It uses the remainingtwo bits to rotate the sequence.This is illustrated in Figure 2.2.1.

An important difference betweenthe sets of spreading sequencesused here and the single Barkercode sequence used for the 1 and2 Mbit/sec rates is that thesesequences are complex. In otherwords, the signal is a (D)QPSKsignal with QPSK spreading.

For all 802.11b bit rates, thepreamble and header are sent atthe 1 Mbit/sec rate. The headeris 192 µsec long (192 bits). Thistranslates to a total of 2112 chips. The payload data is thenappended using one of the fourmodulation rates.

12

Figure 2.2.2. 802.11b PPDU components

Figure 2.2.1. CCK modulation

2.3 Modulation analysis of 802.11b signalsThe 20MHz wide bandwidthof WLAN signals make powerenvelope measurements difficult,as most spectrum analyzers haveresolution bandwidth filters thatare usually limited to less than10 MHz. This means the signal isheavily filtered by the time it getsto the power detector. Peopleoften resort to diode detectorsand oscilloscopes to look at thesesignals, but this approach provideslimited dynamic range, and nofrequency selectivity. Peak-powermeters may provide more dynamicrange, but they generally do not

have bandwidths that are wellmatched to this signal. TheAgilent 89640A VSA has aninformation bandwidth of 36 MHz,making it ideal for time-domainanalysis of WLAN signals. Theinformation bandwidth is thewidest bandwidth that can bedigitized without loss of signalinformation, and is independentof the frequency tuning range.

Shown in Figure 2.3.1 is a plot ofa power-versus-time measurementof a Brand-X Wi-Fi certifiedmodem purchased on the openmarket. Several things areimmediately noticeable fromthe plot. The most obvious is

the transition from the BPSKpreamble/header to the higherrate modulation at the end of theburst. The transition is obviousbecause the QPSK signal has fewchip transitions which cause thecarrier to pass through the origin.The origin in a polar plotrepresents zero carrier power.In the plot, band-power markers,indicated by the solid verticallines, are 192 µsec apart.According to the standard, thisshould be the entire length of thepreamble/header. It appears thatthe power amplifier or DSP iscoming on late and thus shorteningthe sync portion of the preamble.

13

Figure 2.3.1. Power envelope of an 802.11b signal, showing preamble and header

Figure 2.3.2 shows the next burstout of the modem. Both burstswere captured in the analyzer’stime-capture memory. This burst,which has a longer data segment,also comes much closer to havinga full-length preamble.

The measurement shown in Figure2.3.3 is interesting because thepower-envelope display in the

lower grid indicates that the poweris coming on in two steps. Notshown in the plot is the fact that thefirst step occurred 192 µsec fromthe end of the preamble/header.

The Agilent 89640A VSA has theability to make gated spectrummeasurements. The gate intervaland position can be adjustedbefore or after the measurement

data is acquired. With gatemarkers positioned over thesync portion of the burst, it isobvious that the spectrum doesnot exhibit a smooth sin(x)/xshape. This is normal and iscaused by the relatively shortBarker sequence. The spacingof the ripple will generally havesome multiple of 0.5 MHz andis data dependent.

14

Figure 2.3.3. Gated spectrum analysis of power-up of an 802.11b signal

Figure 2.3.2. 802.11b burst with longer data segment

Without taking a new measure-ment, the gate markers can bemoved to analyze the first step inthe turn-on transient. It can beseen in Figure 2.3.4 that the firststep is not modulated. The mostlikely cause of this is the poweramplifier being turned on beforethe baseband signal processing.This is likely because the gatedspectrum, shown in the top halfof the figure, reveals a singletone – carrier leakage – and nomodulation. With this information,it is known that the poweramplifier is coming on at theright time, so the DSP isshortening the sync portionof the burst.

Based on the (spectral) bandpower measurement in theprevious figure, the average signalpower is -38.75 dBm in this over-the-air measurement. The markeron the carrier leakage is at –72.5 dBm, making the carrierleakage about –34 dBc. Thisnumber will come up againduring signal demodulation.

15

Figure 2.3.4. Investigating anomalies in a power burst through gated spectrum analysis

So how does this signal performrelative to the 802.11b mask?Since the mask is drawn on alinear scale, one needs to look atthe signal on a linear scale. Theupper grid in Figure 2.3.5 is anexpanded view of what was seen in Figure 2.3.4 shown inlogarithmic units. The lower traceis exactly the same except that it is displayed in linear units.

One problem with thismeasurement, as described inthe 802.11b standard, is that thelimits do not take into accountdata modulation. This implies a

special test mode. For this signal,the shaping appears to be done inDSP. When special test modes areused to test the pulse shape thereis always the possibility that the"special test software" couldproduce a result that is differentthan the "real software". Figure2.3.5 still provides a good idea ofhow well the pulse is formed,even without a test mode.

16

Figure 2.3.5. 802.11b power-up-envelope mask

The same carrier leakage term isevident when the signal turns off.It is interesting to note that thepulse shaping performed at thebeginning of the burst doesn’tappear at the end of the burst.This could cause a problem becausefast transitions cause spectralsplatter. It does not matter if itis a transition from off-to-on, oron-to-off. In turn-on and turn-offtransitions, the standard does notdictate the exact location of themask. This is rather unusual,and could cause compatibilityproblems.

The 802.11b standard does notspecify a transmit filter. However,filtering is implied in the spectralmask, as seen in Figure 2.3.7.The sin(x)/x plot shown with thespectral mask seems to indicatethat the transmit filterapproximates a RECT function,but that additional filtering isused to limit the sidelobes.

Today, where most of the signalprocessing is done in DSP, manydesigners will probably opt to usea modified RECT baseband filterto control the sidelobe levels. Forexample, a BT=0.5 Gaussian filterwould produce an acceptablespectrum. The alternative is touse RECT filtered signals (i.e. logicsignals) to modulate a carrier andlimit the bandwidth with an IFfilter.

One problem with the 802.11bstandard is that the spectrum maskshown is that of a continuoussignal. To measure the spectrum,the transmitter must be modifiedto produce a continuous signal orsome form of gated spectrumanalysis must be performed onthe burst signal. The problemwith configuring a transmitter forcontinuous output is that the finalresult may not be representative ofactual performance under burstconditions.

17

Figure 2.3.6. 802.11b power-down-envelope mask

Figure 2.3.7. 802.11b transmit spectral mask

Many RF spectrum analyzers havea feature known as gated sweep.The purpose of gating, in all itsvarious forms, is to measure thesignal only when it is present.With swept analyzers, gatedmeasurements can be somewhatdifficult to set up. Most analyzersrequire a separate gate signal totell the analyzer when to measure.Also, the time waveform cannotbe observed during the gatedspectrum measurement to ensurethe gate is properly aligned.

In the Agilent 89640A VSA,spectrums are computed usingFFT’s instead of using swept LO’s,mixers and power detectors. Thisapproach has many advantages,especially when it comes totroubleshooting. Measurementsetup is greatly simplified becausethe Agilent 89640A VSA candisplay the burst signal and thegate alignment while checkingfor spectral mask compliance.

The one problem with the FFTapproach is that the FFT bandwidthis not wide enough for measuringthe entire spectral mask in asingle measurement. However, byadjusting the analyzer’s centerfrequency, the measurement canbe performed in two steps. InFigure 2.3.8, the lower half of thespectral mask is shown (handdrawn) on the spectrum plot.Of slight interest is the spur at an11 MHz offset from the carrier.For strict adherence to thestandard, a swept spectrumanalyzer should be used.

18

Figure 2.3.8. Signal analysis through gated sweep measurements

This spectrogram display inFigure 2.3.9 is comprised ofhundreds of spectrum measure-ments taken over a single burst.Each spectrum measurement isflattened to one row of pixels.Since height can no longer be usedto represent amplitude, color orshade depth is used instead.The spectrogram display is usefulfor looking at the time-varyingspectral characteristics of a burst.For example, the previous figureshowed that the transmitterturned off quickly. This plotshows that the transient is notcausing significant problems. If ithad, there would be a widening ofthe spectrum at the bottom of thespectrogram.

Spectrogram displays have severalattributes:

• Darker shades represent a highsignal level, lighter shades represent a low level.

• The (horizontal) frequency axisis the same as for a regular spectrum display. The vertical axis is time instead of amplitude.The top of the spectrogram traceis the start of the burst and the bottom is the end of the burst.

• A special spectrogram marker,indicated by the horizontal line, was used to select one spectrummeasurement for viewing. This spectrum is shown in the top portion of the lower trace.

• The spectrogram indicatesthat more energy is present at higher frequencies during the BPSK portion of the signal than during the QPSK version. This could possibly be due to cross-over distortion. The power envelope of the burst signal is shown for reference in the bottom of the lower grid.

19

Figure 2.3.9. Spectrogram analysis of a single 802.11b burst

Often in the R&D phase of aproject, the problem is not somuch with the physical layer asit is with the MAC layer. While the Agilent 89640A VSA doesnot directly perform MAC layertesting, it can be used to gaininsight into what’s going on.One of the most powerful featuresof the analyzer is its optional768 Msample deep-capturememory. This memory can beused to capture the signals upto 36 MHz wide without lossof information. At 36 MHz,768 Msamples equates to 8 secof continuous, gap-free data.With the span set to 18 MHz, thememory lasts for 16 sec. With thestandard 12.7 Msample capturememory, the length would be132 msec at 36 MHz.

Much can be learned simply bylooking at the time waveform, asshown in Figure 2.3.10. Here, bothPCMCIA and access point (AP)signals are visible. The last part ofeach burst in the figure revealsthat the AP is using a QPSKmodulation, whereas the PCMCIAcard is staying at BPSK. One burstis significantly longer than theother, and the various timingrelationships between the twotransmitters can easily bedetermined.

20

Figure 2.3.10. Analysis of an exchange between two WLAN modems

Once the signal has been capturedin the analyzer’s memory, it canbe analyzed using a variety oftools. One of the most usefultools is modulation analysis.

As mentioned earlier, thepreamble/header is BPSKmodulated at 1 Mbit/sec.Relative to the QPSK modulationof the payload data, the BPSKmodulation occupies two of thefour corners of a QPSK vectordiagram (Figure 2.3.11). Thedemodulation started before theamplifier was up to full power, ascan be seen in the lower left trace,and in the constellation "spiral"in the center of the constellation.The upper right trace shows thephase error created by instabilityin the LO as the signal is turnedon. Keep in mind that this is not ahopping signal.

The sync-search capability of theanalyzer was used to highlightone symbol, 11 chips, or in thiscase 22 bits (due to the use ofthe 2-bit/sym generic QPSKdemodulator). The single markeris at the same point in time in allfour traces, and it is also onesymbol back from the highlightedsymbol. This measurement canverify that the Barker sequencehas been correctly coded in thebaseband processing. Because ofthe inherent coding gain in DSSS,a single chip error in thissequence would not break theradio; it would just degrade theperformance.

In the summary table aremeasurements of EVM, magnitudeerror, phase error, frequencyerror/offset and carrier leakage.Remember the –-34 dB value forleakage that was estimated usinggated spectrum measurementsin Figure 2.3.4? Here, the valuewas more accurately estimatedat –36 dB. The greater accuracycomes in part from the measurementalgorithms, but mostly becausethe value was estimated over1.4 msec, instead of over just afew microseconds.

21

Figure 2.3.11. Demodulation of an 802.11b preamble

The 802.11b standard uses a metriccalled error vector magnitude(EVM) as a measure of modulationquality. The basic signal errormodel is given in Figure 2.3.12.This metric has become anindustry standard for a widevariety of applications fromcellular phones to cable television.The basic idea behind EVM is thatany impaired signal (usuallycomplex signals) can be representedas the sum of an ideal signal andan error signal. Since the errorsignal cannot be measureddirectly, the test instrumentationreconstructs the ideal signal basedon the detected data and subtractsit from the actual signal todetermine the error signal.

The error signal includes all ofthe following sources of error:

- Additive noise- Nonlinear distortion- Linear distortion

(equalized in 802.11a)- Phase noise- Spurious signals- Other modulation errors

At any single point in time, theerror signal can be representedas a complex vector reaching fromthe ideal point in the I/Q plane, toactual location. Every chip has itsown error vector. EVM is simplythe RMS over 1000 chips.

The ideal vector represents thetheoretical instantaneous magnitudeand phase of the carrier based onthe known data stream. The actualsignal is modeled as the summationof the ideal signal and an errorsignal, as shown in Figure 2.3.13.All signals are complex.

It is worth pointing out that theAgilent 89640A VSA does notcompute EVM exactly as describedin the 802.11b standard. Asdescribed previously, there is anattempt to properly scale themeasured data and to removecarrier leakage terms beforecomputing EVM. Most radiostandards require that EVM becomputed after carrier frequency,carrier phase, carrier leakage, andgain terms have been chosen tominimize EVM. These parameters

are not orthogonal, and thereforemust be simultaneously estimated,or at least iterated, for minimumEVM. The EVM specification of802.11b is a very generous 35%,which would be unusual for aQPSK signal, but is reasonable forDSSS. It is interesting to notethat in 802.11b there is a separatemeasurement for carrier leakage.They do not use the dc I channeland the dc Q channel valuescomputed as part of the EVMmetric.

22

Figure 2.3.12: Signal error model

Figure 2.3.13. Error vector magnitude (EVM)

The Agilent 89640A VSA’s QPSKdemodulator has an adaptiveequalizer that can be used for twopurposes. First, it can be used tocompensate for linear distortionso that non-linear distortion andspurious errors can easily beidentified and quantified. Second,it can be used to determinemultipath characteristics (to thedegree possible using blindequalization).

The 802.11b specification does notallow for equalization prior tocomputing EVM. This means thatany linear distortion, such asgroup delay distortion in the IF,will increase EVM. When EVM ishigh the equalizer can be used asa diagnostic tool. If use of theequalizer significantly improvesthe EVM result, then the channelfrequency response should bechecked for flatness problems (i.e.

group delay distortion). If it doesnot, then the problem is morelikely related to noise, non-lineardistortion, or spurious error.

The measurement shown inFigure 2.3.14 was made with anAP located about eight feet(non line-of-site) from an antennaconnected to the Agilent 89640AVSA. The lower left plot shows the constellation withoutequalization. The upper rightplot is the estimate of the channelresponse computed from theequalizer coefficients, which areshown in the lower right plot. TheEQ coefficients give an indicationof the delay spread in the signal.

When the equalizer is active,metrics such as EVM (not shown)reflect the errors in the post-equalized signal.

23

Figure 2.3.14. Effects of equalization on signal demodulation

3.1 OFDM signalsThe 802.11a and HIPERLANType 2 standards use a modulationtechnology known as OrthogonalFrequency Division Multiplexing(OFDM). It was originally usedin consumer products inDigital Audio and Digital VideoBroadcasting and ADSL modems.It is now finding its way intobroadband internet accesssystems such as WLANs andPoint-to-Multipoint distributionsystems.

The concepts behind OFDM havebeen around for a long time.However, it has only been withinthe last few years that thebaseband processing has beeninexpensive enough to allow forpractical implementations. OFDMwas first proposed as a way ofdealing with multipath. One of theproblems with single carriermodulations (SCM) is that, in agiven environment, as the symbolrate is increased, the symbolinterval becomes shorter thanthe delay spread. Multi-carriermodulation formats solve thisproblem by decreasing the symbolrate and increasing the number ofcarriers. The basic idea is to take asignal and send it over multiplelow-rate carriers instead of asingle high-rate carrier toeliminate inter-symbol interference(ISI) and then compensate formultipath effects with a simpleequalizer. OFDM is a very flexiblemodulation format in that it canbe easily scaled to meet the needsof a particular application. Forapplications like VOFDM, the lackof ISI also greatly simplifies theimplementation of diversityreception.

The 802.11a and HIPERLANType 2 standards are almostidentical in terms of the physicallayer. The remainder of thisapplication note will concentrateon 802.11a, recognizing that themeasurement concepts areapplicable to both standards.

The RF portion of the 802.11bsignal, while challenging to makesmall and inexpensive, is not verychallenging in terms of bits/Hz.There is a sizable error marginbuilt into the format. The CCKformat has a low peak-to-averageratio, making it easy to amplify,and the complementary codeshave some immunity to phaserotations caused by phase noise.The EVM specification of 35%should be easy to maintain in ahigh-volume production line.

The 802.11a format is morechallenging, as the raw data ratesare increased up to 5 times, withoutincreasing the bandwidth. Thechallenges extend from the basicproblems associated with goingto 5 GHz, to the difficulties ofdesigning stable LO’s and efficientamplifiers for the high peak-to-average, multi-carrier OFDMsignals. The challenges will notbe limited to design. The productswill also be more difficult tointegrate and manufacture.

Two signals are orthogonal in agiven interval if, when multipliedtogether and then integrated overthat interval, the result is zero.TDMA is not normally consideredan orthogonal coding scheme.However, the idea applies if thetime interval is considered to bethe burst width. Over that interval,the other signal is zero, so the dotproduct of the two is zero.

Walsh codes, which are used inCDMA systems, are orthogonal,and are probably the mostcommon form of orthogonalsignaling. For example in IS-95,length 64 Walsh codes provide64 possible code channels.

OFDM is actually very closelyrelated to CDMA. Instead of Walshcodes, the basis functions aresinusoids. In a given period, thesinusoids will be orthogonalprovided there are an integralnumber of cycles. The amplitudeand phase of the sinusoid, whichwill be used to represent symbols,does not affect the orthogonalityproperty. Using sinusoids insteadof Walsh codes produces aspectrum in which it is possible toassociate a carrier frequency witha code channel.

24

3. 802.11a and HIPERLAN Type 2

Figure 3.1.1. Orthogonal signals

In traditional frequency divisionmultiplexing systems, the channelspacing is typically greater thanthe symbol rate. This is done inorder to avoid overlappedspectrums. In OFDM, the carriersare orthogonal and overlapwithout interfering with oneanother. The idea is similar to thatof Nyquist filtered SCM signals.The symbols in a single-carriersystem overlap in the timedomain, but do not interfere withone another because of the symbol(T) spacing of the zero crossings.For OFDM, the carriers have aspectral null at all other carrierfrequencies.

Non-linear distortion andphase noise are the two largestcontributing factors to a lossof orthogonality, creating inter-carrier interference. Poorfrequency estimation in thereceiver is another contributingfactor.

Some advantages of OFDM:

• Increased efficiency due to reduced carrier spacing (orthogonal carriers overlap)

• Equalization is simplified, or eliminated

• Greater resistance to fading.• Data transfer rate can be scaled

to conditions• Single frequency networks are

possible (broadcast application)• Available because of advances

in signal processing horsepower

Some disadvantages of OFDM:

• A higher peak-to-average ratio• Increased sensitivity to phase

noise, timing and frequency offsets

• Greater complexity• More expensive transmitters

and receivers• Efficiency gains reduced by

requirement for guard interval

The two biggest RF problems withOFDM are amplification, due tothe higher crest factor, andfrequency accuracy and stability(phase noise). The amount ofadditional power amplifier (PA)backoff, or headroom, required forOFDM is hotly contested. SomeOFDM proponents believe thatonly 1 to 2 dB are required oversingle carrier modulations.

Others believe that the numberis much higher. As with mostrequirements, it really depends onthe assumptions. The amount ofbackoff is a strong function ofadjacent channel considerations,and to a lesser degree it is afunction of in-channel distortion.

The carrier orthogonality is astrong function of the frequencyaccuracy of the receiver and thephase-noise performance of bothTX and RX. Tighter phase noiserequirements and linear PA’scontribute to greaterimplementation costs.

25

Figure 3.1.2, OFDM carrier separation

In 802.11a, the data carrierscan be BPSK, QPSK, 16 QAM, or64 QAM modulated. In 802.11a,only two modulation formats areused simultaneously – BPSK andone of the previously mentionedformats. It is interesting to notethat the center carrier is not usedin either Digital Audio Broadcast(DAB) or WLAN applications.Removing the carrier slightlyreduces the data capacity, butallows the requirements oncarrier leakage to be relaxed.Table 3.1.1 provides a summary ofthe different modulation formatsused in OFDM systems and theirassociated rate-dependentparameters.

The 802.11a and 802.11bsignals occupy about the samebandwidth. Each has its ownspectral mask to be measuredagainst. As before, the standardspecification describes thismeasurement using a standardswept-spectrum analyzer. If aspecial test mode is not availableto output a continuous signal,then gated spectrum analysismust be used.

The OFDM spectrum is very flatacross the top. As drawn in thestandard, the sidelobes could bemistaken to be third-orderdistortion. In fact, the sidelobestructure of this signal is partof the modulation.

One of the problems with singlecarrier modulations is, in a givenenvironment, the symbol intervalbecomes shorter than the delayspread as the symbol rate isincreased. Multi-carriermodulation formats solve thisproblem by distributing the dataacross multiple lower symbol rate

carriers. The symbol interval foreach of the lower-rate carriers ismade longer compared to thedelay spread. To increaserobustness, should a subset of thecarriers be unusable because ofnulls or interference, the

information is interleavedbetween carriers. The interferingtone shown in Figure 3.1.3 woulddo little damage to the OFDMsignal.

26

Table 3.1.1. Rate-dependent parameters

Figure 3.1.3. Transmit spectrum mask for 802.11a

Figure 3.1.4 shows the realreason for the OFDM spectralcharacteristics shown in Figures3.1.2 and 3.1.3. For a single carrier,one can model the transmittedpulse in the time domain as asinusoid multiplied by a RECTfunction. In the frequency domain,this is simply a sin(x)/x shapeconvolved with an impulse at thecarrier frequency.

The sin(x)/x spectrum has nullsat adjacent carrier frequencies,provided that the sinusoid is onfrequency and has zero bandwidth,and the RECT function is theproper width. The RECT functioncan have the wrong width if theADC/DAC sample rates areincorrect, in either the transmitteror receiver. The zero-widthassumption can be violated byphase noise in either thetransmitter or the receiver.

The OFDM signal is comprised of52 carriers. The carriers at theextreme frequencies are theones that contribute most to thesidelobe structure shown in thespectral mask plot shown inFigure 3.1.3.

Basic knowledge about FFT’sshould make the concepts behindOFDM simple to understand. InFigure 3.1.5, an OFDM signal withonly one carrier is created. Themagnitude and phase of the carrierare determined from the symbolto be transmitted, as shown inthe constellation diagram. Thecomplex number representingthe symbol is loaded into an FFTbuffer, and an inverse-FFT (IFFT)is performed. This produces a setof time-domain samples, whichare then transmitted. In 802.11aand HIPERLAN Type 2, the FFTsize is 64, with 52 of the FFT binsloaded with data and pilots. Afterthe IFFT, all 64 time samples aretransmitted.

While the first pulse is beingtransmitted, the next symbol isloaded into the FFT buffer.Notice that the second pulse(Figure 3.1.6), when joined withthe first, results in a discontinuity.This is normal. The resultingspectral splatter can be attenuatedsomewhat by windowing the data,as described in the 802.11astandard.

27

Figure 3.1.4. Spectral shape of a signal OFDMcarrier

Figure 3.1.5. IFFT used to create one carrier

Figure 3.1.6. One IFFT per symbol period

Figure 3.1.7 shows how a multi-carrier OFDM signal is generated,and how easy it is to have manydifferent modulation formats.As more carriers are added, theresulting time waveform becomesmore complex. This is one of theproblems with OFDM. Describedlater, the addition of multiplecarriers results in a signal with ahigh peak-to-average power ratio.

There are many ways to considerthe orthogonality properties inOFDM. In terms of FFT’s, a signalthat is perfectly periodic in theFFT time record has nulls inadjacent FFT bins.

The reason that OFDM is moresensitive to phase noise shouldnow be obvious. The phase noiseis an additional modulation thatwill modify the sin(x)/x spectrum,reducing the depth of the nulls,and creating interference to othercarriers (not shown).

Receiver frequency trackingis critical. If the receiver is offfrequency, then the nulls in eachcarrier will not land on a FFT bin(Figure 3.1.9). In FFT terminology,this is called leakage. For OFDM,the result is inter-carrierinterference (ICI).

28

Figure 3.1.7. IFFT used to create multiple carriers

Figure 3.1.8. For FFT, nulls are "on bin" if thetone is on bin

Figure 3.1.9. Leakage effects of frequencyerrors and phase noise

3.2 Dealing with multipathThe processing of a receivedOFDM signal is illustrated inFigure 3.2.1. The waveform isdigitized and converted back toa symbol using an FFT. Tu refersto the meaningful part of thewaveform.

The simple OFDM signal generatedin the example will not workunder multipath conditions. Asshown in Figure 3.2.2, if the FFTis aligned with the biggest signal,then the other signal paths willintroduce ISI.

In the example shown earlierin Figure 3.1.6, the effects of thechannel were ignored, and twopulses from subsequent symbolswere spliced together. This willnot work in practice because thechannel will still introduce ISIbetween pulses. To combat theproblem, the pulse is modifiedby a technique know as cyclicextension (see Figure 3.2.3).In this process, the last 1/4 of thepulse is copied and attached tothe beginning of the burst. Thisis called the guard interval (Tg).Due to the periodic nature of theFFT, the junction between theguard interval waveform and thestart of the original burst willalways be continuous. However,the waveform will still suffer fromdiscontinuities at the junctionsbetween adjacent symbols.

29

Figure 3.2.1. Receiving an OFDM signal

Figure 3.2.2. Effects of multipath delay

Figure 3.2.3. Cyclic extension

Figure 3.2.4 illustrates how theaddition of a guard interval helpswith ISI. Shown are two copies ofthe same signal. Each copy takes adifferent path so they arrive at thereceiver at slightly different times,where they are combined in thereceiver’s antenna into a singlesignal. In the time intervaldenoted by the box marked Tu,the signal will only interfere withitself. This amounts to a scalingand rotation of the symbol,nothing more.

In the guard interval region (Tg),it is easy to see that the resultingsignal will have contributions fromboth symbols – ISI. The guardinterval is ignored in the receiver,so that the ISI does not degradereceiver performance. Obviouslythe guard interval needs to belonger than the delay spread, butnot so long that throughput is lost.In the 802.11a standard, the guardinterval is fixed.

The windowing of data to reduceout-of-band power (Figure 3.2.5)involves multiplying the guardinterval region by a shapingfunction, usually a raisedcosine window function. Thedisadvantage of this technique isthat the window interval intrudesinto the next symbol’s guardinterval, thus reducing thatsymbol’s resilience to interferencefrom multipath delay.

Delay spread is not limited topositive delays. In non line-of-siteconditions, the shortest pathmay not be the strongest. Theimplication on OFDM receivers isthat the FFT may not be perfectlyaligned with the useful part of theburst (as it is often called, sincethe guard band is simply discardedby the receiver). Instead, thereceiver will shift the FFT locationto the left, using part of the guardinterval instead of only the entireportion of the useful part of theburst.

30

Figure 3.2.4. Guard interval and signal interactions

Figure 3.2.5. Windowing of OFDM symbols to reduce out-of-band power

3.3 Modulation analysis of an OFDM signal

Figure 3.3.1 shows an 802.11aburst. The sidelobe structurecaused by the sin(x)/x spectrumsof the individual carriers is clearlyvisible in the spectrum plot. Notethat the spectrum is not veryuniform. This is caused in part bythe preambles, but mostly by thedata being transmitted. In thepower-versus-time plot in thelower trace, three distinct regionsof the burst are visible.

An OFDM burst actually has fourdistinct regions (Figure 3.3.2).The first is the Short TrainingSequence, followed by a LongTraining Sequence, and finally bythe SIGNAL and DATA symbols.From an RF standpoint, theSIGNAL symbol and the rest ofthe OFDM symbols are similar.

31

Figure 3.3.1. OFDM power burst and spectrum

Figure 3.3.2. OFDM training structure

The spectrum of the short trainingsymbol can be seen (Figure 3.3.3)by using the gate markers. Thissymbol uses every 4th carrier, sothe carriers are widely spaced.The center carrier is not used in802.11a. The wide carrier spacingof the short training symbol makesthis symbol interval ideal for easycarrier leakage measurements.

When the gate markers are movedover to the long training symbol,we can see that the spectrum isnice and flat (Figure 3.3.4). Forthis symbol, all of the carriers(save the one in the center, whichis not used) have the sameamplitude. The signal flatness iseasily measured using the longtraining symbol interval.

32

Figure 3.3.3. 802.11a short training symbol spectrum

Figure 3.3.4. 802.11a long training symbol spectrum

It is useful to consider OFDMfrom a 2-dimensional standpoint.Figure 3.3.5 shows the shortsequence, which uses every fourthcarrier, followed by the longsequence, and finally by the datacarriers. In 802.11a, four of the52 carriers in the data portion ofthe burst are pilots, although onlyone is shown (darkly shadedcolumn in the middle). The datacarriers change in power level ona symbol-by-symbol basis when16 or 64 QAM is used.

The spectral characteristicsof OFDM signals as a functionof time can be observed using aspectrogram, shown in the tophalf of Figure 3.3.6. In the lowertrace, the spectrum is shown ata point in time indicated by thehorizontal line in the upper trace.The top of the upper tracecorresponds to the beginningof the burst.

The short sequence, using everyfourth carrier, is clearly visible,along with the discrete tones inthe resulting sidebands. Alsovisible is the spectral splatter tothe left and right of the signalcaused by the discontinuitiesbetween symbols. The latter ismost visible at the junctionbetween the short and long syncs.

For reference, the power-versus-time plot is shown in the upperpart of the lower grid.

Given the spectral splatter of thissignal, it would be dangerous tomodel the effects on an adjacentchannel by assuming additivewhite Gaussian noise (AWGN).

33

Figure 3.3.5. 802.11a carrier assignments.

Figure 3.3.6. Spectrogram analysis of an OFDM burst

Figure 3.3.7 shows the 802.11aconstellation for an entire burst.Note this constellation containsboth the BPSK format alwaysused in the training sequences,the signal symbol, and on the pilotcarriers, along with the format (16QAM in this case) on the datacarriers. The standard requires anadaptive equalizer. The lower lefttrace shows the phase response ofthe equalizer used for this burst.Viewing the complex equalizerresult provides a measure oftransmitter performance and/orthe propagation path. The upperright trace shows the EVM of eachcarrier in the burst. Each columnof dots shows the EVM of a carrierfor each symbol in the burst. Thedark line shows the average EVMacross the carriers. The shape ofthis profile indicates problemssuch as gain imbalance and I/Qtime mismatch. The table in thelower right gives some of the keymetrics of this signal includingoverall EVM in dB, anotherrequirement of the 802.11aspecification.

The power envelope of an OFDMburst is not constant. A singlemetric, peak-to-average ratio(PAR), is often used to describethe amount of headroom requiredin an amplifier. For OFDM signals,this metric is not very usefulbecause the "real peak", whateverthat is, may not occur very often.

It is more meaningful for OFDMsignals to associate a percentageprobability with a power level.As shown in the Figure 3.3.8, thesignal exceeds the average power(dark horizontal line) 40% of thetime (it would be 50% only if themean and median were identical).It exceeds a level that is 4 dBabove the average, 5% of the time.In other words, if this particularsignal were run through a PA with4 dB of headroom, or back-off, thesignal would clip 5% of the time.This is more useful than knowingthat the peak-to-average for thissignal is 8 dB.

34

Figure 3.3.8. Power versus time for an OFDM burst

Figure 3.3.7. OFDM constellation and EVM plots

The best way to look at powerstatistics is using the complementarycumulative distribution function(CCDF). In the measurementshown in Figure 3.3.9, the gatemarkers are used to select theactive portion of the burst. If thiswere not done, the periods whenthe burst is off would bias downthe average power calculation.

The CCDF, which is simply themore common CDF subtractedfrom 1.0, shows dB above averagepower on the horizontal axis, andpercent probability on verticalaxis. The marker shows that, forthis one burst, the signal exceeds7.2 dB above average 0.09% ofthe time. Normally, the CCDFmeasurement would be madeover several bursts to improvethe confidence interval on thelow probability peaks. The curvedgraticule line represents thestatistics for Gaussian noise. MostOFDM signals will have statisticsthat follow that line very closely.

The EVM concept used for 802.11bis valid for 802.11a. For 802.11b,there is only one constellation, butfor 802.11a the constellation willchange depending on the datarate. Unlike QPSK, the OFDMconstellations do not have all ofthe symbols at the same amplitude(distance from the origin).

The OFDM constellations arenormalized to unity power. Forthis reason, the EVM computationdoes not need to be normalized –provided the transmitted dataproduces a uniform distributionof constellation points. A morerigorous approach to EVMcomputation would not assumea uniform distribution oftransmitted symbols and wouldinstead factor in the averagepower level of the ideal symbolsactually transmitted.

35

Figure 3.3.9. CCDF curve

Figure 3.3.10 shows an OFDMconstellation in the upper trace.It is a composite plot of allcarriers over all symbols. Theconstellation is a composite of theBPSK carriers and the 16 QAMcarriers. Shown in the lower traceis the adaptive equalizer response.In 802.11a, the long trainingsymbol is used to train anequalizer. This is a requiredstep in the measurement process.The complex equalizer result canbe viewed as magnitude, phase orgroup delay. The magnitude andphase responses are shown. Theequalizer result is a good measureof transmitted flatness and carrierpower levels.

The 802.11a standard establishesRelative Constellation Errors foreach of the data rates. These areexpressed in dB, while the errorcomputation is defined in RMSunits. The allowable EVM valuesin %RMS are shown in Table 3.3.1.

In 802.11a, data rates of 6, 12 and24 Mbit/sec are mandatory forcompliance with the standard,and utilization of the remainingdata rates in the specificationis optional. In a productionenvironment, it should benecessary to measure the EVMonly at the highest rate supported.Outside of slightly different powerstatistics, there are very few errorinducing mechanisms that wouldcause a transmitter to have asignificantly different measuredEVM for each rate (given thenormalized constellations).

The highest mandatory data rateis 24 Mbit/sec, for which an EVMof 15.8% is specified. 54 Mbit/secmodems will need to achieve5.6% EVM. It may be that somecompanies plan on gradingmodems as they come off of theproduction line. Modems thatachieve 5.6% EVM could be labeledas 54 Mbit/sec capable modems,while those that have worseperformance could be sold as24 Mbit/sec modems, insteadof being scrapped.

36

Table 3.3.1. 802.11a modulation quality and relative constellation error

Figure 3.3.10. EVM for OFDM

This application note has examinedthe modulation technology andsignal characteristics behind themajor WLAN standards, and therelevant signal measurement andanalysis techniques needed inorder to evaluate RF performance.The principal focus of thisdocument has been on 802.11b,802.11a, and HIPERLAN Type 2.

Testing is an important part ofthe design and manufacturingprocess. This is especially true in WLAN systems wheninteroperability is a goal.Although the application note has mostly considered testing ofthe transmitted signal, receivertesting is equally important.What may not be obvious is thatmost of the measurementsdiscussed in this paper are veryimportant for receiver testing andtroubleshooting. For example, if a receiver fails a sensitivity test, it could be because of thereceiver, a problem with the signal used to perform the test, or other factors.

Measurement tools are availableto speed the development processby providing diagnostic capabilities,which can help to quickly isolatethe source of a problem, whetherit is in the transmitter or receiver,at RF or baseband. An Agilent89600 Series VSA can measuresignal quality at many points inthe system. These points includebaseband (I/Q), IF, and RF, inboth the transmitter up-conversion signal path and thereceiver down-conversion signalpath. If there is a problem with areceiver and there is confidencein the quality of the signal goinginto the receiver, then it wouldbe useful to make many of themeasurements described in thispaper at test points within thereceiver’s signal path. Onesuch example is a spectrummeasurement at the output of ananalog IQ demodulator, just beforethe receiver digitizes the signal.Using a 2-channel Agilent 89610VSA in ch1+jch2 mode, thebaseband signal can be checkedfor spurious error, adjacentchannel leakage, receiver-inducedphase noise, EVM, etc. A single RFchannel Agilent 89640 VSA can beused at IF and RF frequencies. Alternatively, you can use the89600 VSA software with otherfront ends, such as spectrumanalyzers, Infiniium oscilloscopes,and even Agilent logic analyzersfor complete and in-depth analysisanywhere in your equipment’sblock diagram.

37

4. Conclusion

(D)BPSK – differential binary phase shift keying

(D)QPSK – differential quadrature phase shift keying

ADC – analog-to-digital converter

Agilent ADS – Agilent advanced design system

ADSL – asynchronous digital subscriber line

AP – access point

AWGN – additive white Gaussian noise

BPSK – binary phase shift keying

CCDF – complementary cumulative distribution function

CCK – complementary code keying

CDF – cumulative distribution function

CDMA – code division multiple access

CRC – cyclic redundancy code

DAB – digital audio broadcast

DAC – digital-to-analog converter

DSP – digital signal processing

DSSS – direct sequence spread spectrum

EIRP – effective isotropic radiated power

EQ – equalizer

ESG – electronic signal generator

ETSI – European Telecommunications Standards Institute

EVM – error vector magnitude

FFT – fast Fourier transform

FHSS – frequency hopping spreadspectrum

FM – frequency modulation

FSK – frequency shift keying

GFSK – Gaussian frequency shift keying

GMSK – Gaussian minimum shift keying

GSM – global system for mobile communications

HIPERLAN – high performance radio local area network

I/Q – in-phase/quadrature

ICI – inter-carrier interference

IF – intermediate frequency

IFFT – inverse fast fourier transform

IS-95 – cdmaOne cellular phone communications standard

ISI – inter-symbol interference

ISM – industrial, scientific and medical band

LAN – local area network

LO – local oscillator

MAC – medium access control

MPDU – MAC sublayer protocol data units

MSK – minimum shift keying

OFDM – orthogonal frequency division multiplexing

PA – power amplifier

PAR – peak-to-average ratio

PCMCIA – Personal Computer Memory Card International Association

PLCP – physical layer convergence procedure

PMD – physical medium dependent

PPDU – PHY protocol data units

PSDU – PLCP service data units

PSK – phase shift keying

QAM – quadrature amplitude modulation

QPSK – quadrature phase shift keying

RF – radio frequency

RMS – root mean square

RX – receiver

SCM – single carrier modulation

SFD – start frame delimiter

TDMA – time division multiple access

TX – transmitter

VOFDM – vector orthogonal frequency division multiplexing

Wi-Fi – 802.11b certification

WLAN – wireless local area network

38

Acronym Glossary

Related LiteratureAgilent 89600 Series Vector Signal Analyzers

Configuration Guide, literature number 5968-9350E

Agilent 89600 Vector Signal Analysis, Technical Overview, literature number 5989-1679EN

Digital Modulation in Communication Systems-An Introduction, Application Note 1298, literature number 5965-7160E

Testing and Troubleshooting Digital RF Communications Transmitter Designs, Application Note 1313, literature number 5968-3578E

Testing and Troubleshooting Digital RF Communications Receiver Designs, Application Note 1314, literature number 5968-3579E

Using Vector Modulation Analysis in the Integration, Troubleshooting and Design of Digital RF Communications Systems,

Product Note, literature number 5091-8687E

Using Error Vector Magnitude Measurements to Analyze and Troubleshoot Vector-Modulated Signals, Product Note, literature number 5965-2898E

10 Steps to a Perfect Digital Demodulation Measurement, Product Note, literature number 5966-0444E

39

WLAN Standards Summary Table

WLAN standards

802.11 802.11b 802.11a HIPERLAN type 1 HIPERLAN type 2

Frequency band 2.4 GHz 2.4 GHz 5 GHz 5 GHz 5 GHz

Channel separation 25 MHz for DSSS, 25 MHz 20 MHz 23.5 MHz 20 MHz1 MHz for FHSS

Maximum raw data rate 2 Mbit/s 11 Mbit/s 54 Mbit/s 23.5 Mbit/s 54 Mbit/s

Carrier type FHSS or DSSS DSSS OFDM single carrier OFDM

Modulation GFSK (FHSS), CCK BPSK & QPSK, FSK or GMSK BPSK & QPSK,DBPSK or DQPSK 16 QAM, or 16 QAM, or 64 QAM

(DSSS) 64 QAM

Number of carriers 792 1 (DSSS) 48 data & 4 pilot 1 48 data & 4 pilotper channel

Maximum power output 30 dBm 30 dBm 35 dBm1 30 dBm1 30 dBm1

1 Effective Isotropic Radiated Power (EIRP)2 Indicates number of operating channels for FHSS in USA

Agilent Email Updates

Agilent Direct

AgilentOpen

Remove all doubt

Our repair and calibration services will getyour equipment back to you, performinglike new, when promised. You will get full value out of your Agilent equipmentthroughout its lifetime. Your equipment willbe serviced by Agilent-trained techniciansusing the latest factory calibrationprocedures, automated repair diagnosticsand genuine parts. You will always have theutmost confidence in your measurements.

Agilent offers a wide range of additionalexpert test and measurement services foryour equipment, including initial start-upassistance onsite education and training,as well as design, system integration, andproject management.

For more information on repair andcalibration services, go to

www.agilent.com/find/removealldoubt

www.agilent.com

For more information on Agilent

Technologies’ products, applications

or services, please contact your local

Agilent office. The complete list is

available at:

www.agilent.com/find/contactus

Phone or Fax

United States:(tel) 800 829 4444(fax) 800 829 4433

Canada:(tel) 877 894 4414(fax) 800 746 4866

China:(tel) 800 810 0189(fax) 800 820 2816

Europe:(tel) 31 20 547 2111

Japan:(tel) (81) 426 56 7832(fax) (81) 426 56 7840

Korea:(tel) (080) 769 0800(fax) (080) 769 0900

Latin America:(tel) (305) 269 7500

Taiwan:(tel) 0800 047 866 (fax) 0800 286 331

Other Asia Pacific Countries:(tel) (65) 6375 8100 (fax) (65) 6755 0042Email: tm_ap@agilent.comRevised: 11/08/06

Product specifications and descriptions

in this document subject to change

without notice.

© Agilent Technologies, Inc. 2001, 2007

Printed in the USA January 10, 2007

5988-3762EN

www.agilent.com/find/emailupdatesGet the latest information on the productsand applications you select.

www.agilent.com/find/agilentdirectQuickly choose and use your test equipment solutions with confidence.

www.agilent.com/find/openAgilent Open simplifies the process ofconnecting and programming test systemsto help engineers design, validate andmanufacture electronic products. Agilentoffers open connectivity for a broad rangeof system-ready instruments, open industrysoftware, PC-standard I/O and globalsupport, which are combined to moreeasily integrate test system development.

is the US registered trademark of the LXI Consortium.

To stay up to date on WLAN andBluetooth™, please visit

www.agilent.com/find/wlan

Bluetooth and the Bluetooth logos are trademarks

owned by the Bluetooth SIG, Inc., U.S.A. and licensed

to Agilent Technologies, Inc.