The harmful effects of local oscillator noisemillimeterwave.free.fr/F9HX lo noise.pdfTo test phase modulation, an 8038 deliv-Fig 5: Spectral density of a very low noise OCXO spoiled

1Introduction

The goal of any transmission system isto transfer information from point topoint. It is correctly done when thereceived information is identical to theoriginal one. In practice, the informa-tion can be distorted and/or mixedwith disturbing signals. Noise of alltransmission stages is the main trou-blemaker. At the limit, they can be sohigh that the ratio information/noiseratio is too low to get the informationavailable.

2Local oscillator

A local oscillator, LO, is composed of anoscillator eventually followed by mixersand/or frequency multipliers in order toobtain the desired frequency, because itis not possible or desirable to start with avery high frequency oscillator. As with all electronic devices, an oscilla-tor makes noise. As long as oscillatorswere based on transistor and crystalarrangement, their noise was acceptable.The use of lower Q resonators and the

modern devices PLL and DDS arequickly showing their weakness wheremore demanding performances are re-quired. Noise level is an important limitatingparameter for weak signals communica-tions as it decreases the RX sensitivityand TX signal quality.More, other shortcoming can upset weaksignal reception when strong signals areon adjacent channels: it is called recipro-cal mixing. Those defects are restricting for bothamateur and professional equipment.

3Electronic circuits noise

Electronic circuits are composed of ac-tive and passive components. It is wellknown that any conductor generatesnoise by thermal agitation of electrons,called white noise as it extends from DCup to the highest frequencies. The power of thermal noise is given by:

Pnoise = k T B (watts)where:

k = Boltzmann’s constant (1.38.10-23

K/J)

André Jamet, F9HX

The harmful effects of localoscillator noise

VHF COMMUNICATIONS 3/2008

130

T = absolute temperature (K = 273°C + ambient)

B = bandwidth (hertz)At 27°C and one hertz of bandwidth, wehave:

Pnoise = 1.38 x 10–23 x 300 = 4.14 x 10-21 (watt) = 4.14 x 10-18 (mW) = 10 log (4.14 x 10-18) = -174dBm/Hz

(see Appendix)Oscillators produce other kinds of noise.Most influential are:

• Shot noise occurs when there is apotential barrier, i.e. diodes and tran-sistors junctions

• Flicker noise, called pink noise or1/f noise as it is decreasing with thefrequency up to a cut off frequencyand remains constant after that. Me-tallic resistors produce the lowestand it is more prominent in FET andMOS transistors.

Powers of different noises add.

4Oscillator noise

An oscillator is composed of an amplifierin conjunction with a high Q resonator.Oscillator behaviour survey comprises:Whys: various noise voltages and cur-rents are present together with normalDC and HF ones. They are due to variousoscillator components, passive and ac-tive. Wherefores: those voltages and currentsmodulate the normal ones by amplitudeand phase modulation. Mixer and other stages following theoscillator attenuate amplitude-modulatednoise. Therefore, amplitude noise is gen-erally treated as immeasurable and ne-

glected. If it were not, any residual AMnoise would produce a familiar effect. Onthe contrary, phase-modulated noise pro-duces less intuitive effects and requiresmore attention to be well understood.Even after saturated stages phase modu-lation remains. Moreover, the phase off-set is multiplied by any multiplier stage.Usually, we use the expression: phasenoise.

5Low Noise LO design

To design a low noise oscillator, we needto follow some requirements. It is the aimof this article, to minimise the phasen o i s e , t h e a d v i c e o f t h eKA2WEU/DJ2LR, Pr. Dr Ulrich Rohdecheck list [5,6] and F5RCT lecture [7] is:

• Maximise the unloaded resonator Q

• Maximise the resonator reactive en-ergy

• Avoid oscillator saturation

• Choose an active device with thelowest possible noise figure at theactual working conditions, espe-cially for flicker noise under 10kHz

• Use low frequency negative feed-back to reduce transistor flickernoise

• The energy should be coupled fromthe resonator rather than another partof the oscillator

• Use high stability and low noisepassive components. Be careful withlow Q varicap diodes.

More, external reasons can add noise:

• DC supplies from linear and switch-ing regulators

• Vibrations (fan) and mechanical


131

shocks (microphonic effect of reso-nators and some capacitors andcoils)

• Digital circuits (computer).

6Whole oscillator noise

When multipliers stages follow an oscil-lator the phase noise is increased by themultiplication factor. For an N factor,noise is degraded by 20 log N. In realworld conditions, it is a minimum valueincreased by the multipliers stages ownnoise. Therefore, a choice should bemade between:

• To start from a rather low frequencyoscillator that requires a large multi-plication

• To start from a high frequency oscil-lator.

On one hand, high stability, low retraceeffect and low noise sources, such asOCXOs, caesium or rubidium generators,usually generate a 10MHz signal. A verylarge multiplication factor is required forthe microwave bands. On the other hand,VHF OCXOs are able to give an accept-able solution if the drift and the retraceeffect are cancelled by a permanent

power supply [8].A 10MHz OCXO can be used as areference for PLL and DDS devices [9].A microwave PLL can be based on aDRO.Those solutions avoid a large multiplica-tion factor, but they are more difficult toachieve.

7Real oscillator spectrum

Fig 1 shows the frequency spectrum ofan ideal and real oscillators. The spec-trum spread is due to the phase noise.This is common for analogue signals,like those from an LO. Fig 2 shows thetime domain. In this case, the word jitteris used in digital applications. From the Fig 1, we can see two side-bands around the central frequency calledthe carrier. Fig 3 shows the completespectrum (DSB) and the half-sidebandspectrum (SSB). As the phase noisespectrum is symmetric around the carrier,it is commonly characterised by the SSBspectrum. The symbol is Lϕ (fm). It repre-sents the ratio of the SSB noise power ina one hertz bandwidth centred at fm hertzaway from the carrier, to the carrierpower. It is called SSB spectral density;

Fig 1: Frequencyspectrum of anideal and a realoscillator.

Fig 2: Phase noisein the time domain(jitter).


132

the units are dBc/Hz.If we assume that the phase noise isapproximately constant over the band-width of interest, at a distance fm fromthe carrier and for a bandwidth B, noisepower is given by:

Pin B hertz = B Lϕ(fm) Pc (watt)where Pc is the carrier power. It is generally more convenient to workin dBs. Then:

Pin B hertz = 10 log(B) + Lϕ(fm ) + Pc (dBm)

That offset (fm) can be 10Hz, 100Hz,1kHz, 10kHz or 100kHz.Spectral density can be calculated fromLeeson [1] who established a formulausing the oscillator’s parameters. Anumber of surveys [2,3,4] added correc-tions and complements.Fig 4 shows the general look of acalculated curve. We can see several

regions with different slopes. Near thecarrier, up to a few tens of hertz, theslope is following an f-3 law, so a -30dB/decade slope. Within the kilohertzregion, the law is f-2 equals to -20dB/decade. Then, follows the thermalnoise region with a constant value. A more detailed analysis shows otherregions in f-4 and f-1. After reading ofseveral surveys, it is questionable, so theabove is mentioned.

8LO noise measurement

To measure an LO noise spectral density,we need equipment able to handle mega-hertz and even gigahertz signals. In gen-eral, its own noise near the carrier wouldbe at the same level or even more thanwe want to measure.

Fig 3: DSB andSSB spectrums.

Fig 4: Simplifiedtheoretical curve ofthe noise spectraldensity of anoscillator.


133

Specialised equipment is not common fora radio amateur. For example, the E 5052from Agilent cost is $85,000! The Aer-oflex PN 9000 is able to do the test at aquite high cost. An affordable method is to use a veryclean auxiliary signal, a detector or aphase discriminator to get a signal ac-ceptable for use with standard equipment Fig 5 shows the spectral density of a verylow noise OCXO. Nevertheless, we cansee large undesirable spurious signalsbetween 100 and 1000Hz. They comefrom the power supply (main frequencyharmonics and variations). However, we can show a poor qualityoscillators spectrum with a commonspectrum analyser. That will be detailedlater with added noise oscillators. Moreo-ver, we can listen an oscillator signalwith a receiver to “hear” its noise but not

be able to quantify the noise. That is veryuseful for UHF and microwaves.

9Test of deliberately added noiseoscillators

It is useful to know the effect of a noisepolluting a sine wave signal. To test amplitude modulation effect, a1000Hz signal from an audio generator ismixed using a dual gate FET (BF 988)with noise. The noise is produced by adevice comprising of a 78L09 regulatorfollowed by two stages of amplification[11]. On an oscilloscope, we can dis-tinctly see the noise above 1% modula-tion. With headphones, we can hearmuch lower levels.To test phase modulation, an 8038 deliv-

Fig 5: Spectraldensity of a verylow noise OCXOspoiled by a poorpower supply.

Fig 6: Narrow frequency modulationdevice.

1039 9,600 -23 50 1058 9,560 -4 10 1062 9,550 0 0 1080 9,510 18 -401103 9,460 41 -90

F U8 Δf Udc

(Hz) (mV)

Table 1: Fig 6 calibration.


134

ers a 100Hz signal with a narrow fre-quency modulation (NBFM produces thesame effect as phase modulation). Fig 6shows the block diagram. First, the de-vice is calibrated with a DC signal (seeTable 1). We have:

Δf / ΔUdc (average) = 1Hz / 2.2 mVdc

NB: there is no amplitude modulation, asthe amplitude remains constant. The noise generator already described isused with this modulation. The thresholdis 40mVpp that is to say about 18Hz or a1.8% frequency variation. NB : the audio effect is not the samewhen the signal is amplitude or phasemodulated by the noise. It is difficult toexplain this! It seems deeper and moredisturbing even when the bandwidths are

the same.

10TX signal purity versus LOnoise

LO noise spoils TX signals even whenthere is no mixer, only frequency multi-pliers. The received signal spreads, as alarge part of the power is lost in undesir-able sidebands. The noise appears aroundthe received carrier. It is really a jammer!F1GHB and F5EFD “listened” to twoLOs at 10GHz. One is an OCXO fol-lowed by multipliers. Its measured noiseis about -90dBc/Hz. The other is asynthesiser based on a LMX 2326 havingnoise of about -63dBc/Hz. With the firstone, nothing was wrong but audible noisewas heard on the second. Radio amateurs working the microwavebands above 76GHz know how difficultit is to obtain a pure note. SSB and CWcan be upset by the phase noise.WA1ZMS has made a presentation atMUD 2004 [10]; we can “hear” noise ofseveral oscillators at 241GHz.

11Noisy HF TX

To known how noise upsets a HF TX, thedevice in Fig 7 is able to produce a noisy20MHz signal. This device comprises a HF generator(ADRET 6315) to deliver an 80MHzsignal. The noise generator can modulatethat generator. A mixer mixes the outputsignal and a 100MHz produced by anOCXO. A spectrum analyser shows theoutput at 20MHz. Scan is onekHz/division, bandwidth 100Hz andscale 10dB/division.

Fig 7: HF device to produce a noisyHF signal.

Fig 8: Original 20MHz spectrum.


135

Figs 8 and 9 show the 20MHz spectrumsbefore and after noise is added. The floornoise, noise without signal, increases bymore than 15dB. With a RX in AMmode, we can hear the noise on each sideof the tuning frequency, as we hear aNBFM signal. In NBFM mode, noise isat the centre frequency. In SSB mode, wecan find the noise on one side. For thosethree cases, listening seems not to beaffected by this noise level.

12Deliberately noisy microwavesignal

Fig 10 shows the equipment. It startswith a XO at 108MHz, phase modulatedby a varicap. The noise generator is as

above. Several multipliers stages followto produce a 10,368MHz signal. Becauseof the large multiplication factor and theheavy saturation, the output signal iseasily phase modulated without notice-able AM. With a 10GHz RX, we can hear thesound effect. In SSB mode without addednoise, tuning gives the usual tone varia-tion, low to high frequency. With anincrease added noise the tone become“rougher and rougher” My “poor mans lab” does not allow meto quantify precisely that effect. Never-theless, my old 141T spectrum analyserwith an 18GHz plug-in is able to showme the spectrum variations during thetest. The floor noise of that old spectrumanalyser is -110dBm for a 100Hz band-width in the 8.23 to 14.35GHz range.Figs 11 and 12 clearly show the noiseeffect. We can see that the reports byF1GHB and F5EFD are roughly con-firmed. A 20dB of noise increasechanges a signal from correct to unac-ceptable.By interpolation, we have approximately:Fig 11

• -50dBc at 1kHz offset; then:

• -50 -log 100 (B in Hz) = -70dBc/Hz

Fig 12


• -30 -log 100 = -50dBc/Hz

Fig 9: 20MHz spectrum with addednoise.

Fig 10: Equipment to produce a noisy microwave signal.


136

13Noise LO effect on a RXsensitivity

The 144MHz transceiver [F9HX 12] isbased on an IQ mixer and a zero interme-diate frequency. A 24MHz VFO with afrequency variation by a varicap is fol-lowed by multipliers stages to get an LOoutput at 144MHz. Figs 13 and 14 showthe 144MHz spectrum before and afterdeliberately added noise. Those spectrums are displayed with a100Hz bandwidth, a 1kHz/division scan

and 10dB/division. We obtain approxi-mately:Before added noise:


• -80 - log 100 = -100dBc/HzAfter added noise:

• -40dBc/Hz at 1kHz offset; then:

• -40 - log 100 = -60dBc/HzIn the last case, in SSB mode, signals areheavily spoiled. Weak signals are unin-telligible.

Fig 11: Correct 10,368MHz spectrum,BW=100Hz, scan=5kHz/division, scale=10dB/division.

Fig 12: Noisy 10.368MHz spectrum,BW=100Hz, sacn-5kHz/division scale=10dB/division.

Fig 13: Correct 144.300MHzspectrum.

Fig 14: Noisy 144.300MHz spectrum.


137

14Reciprocal mixing

That is the second LO noise effect. Itlimits the low signals reception in thepresence of loud adjacent signals (Fig15). When the LO noise mixes with aninterfering signal, a signal is produced inthe IF bandwidth B. It can perturb oreven damps it. This effect is very awkward on the HFand VHF bands during the contests. Aloud station can be near a weak one. TheRX seems to go “deaf”. From [13 and others], we can calculatethe reciprocal mixing effect. For exam-ple, a 2m band RX with a 2700Hzbandwidth 9MHz IF and a -90dBc/Hznoise LO. It is tuned for a S3 = -129dBmsignal. An S9+20dB = -73dBm interfer-ing station is 10kHz away. Approximate calculation gives:

• LO noise power in 270Hz: - 90 + 10log 2700 = -90 + 34 = -56dBm

• Mixer output from LO noise + inter-fering signal = -73 - 56 = -129dBm.

We notice that the noise level producedby the interfering signal equals the de-sired signal. A test verifies that calculation. Two RFgenerators feed simultaneously a RXwhile the LO is added noise or not.Without, a weak signal is audible; with, itis almost lost in noise.

15Conclusion

As we can see, LO noise is an importantparameter for communication equip-ments. In the HF and VHF bands, it canupset weak signals reception by loudadjacent signals; that is currently the caseduring contests. For very low signals communications, itis a limiting factor. Microwaves are verydemanding owing to the very large multi-plication factor in LO. For any more information contact: [email protected]

16Appendix

Thermal noise comprises two orthogonalcomponents not correlated: amplitudeand phase. Each is half of the total.Consequently, from [18]:

Pphase noise = Pamplitude noise = -174 - 3 =-177dBm/Hz

Nevertheless, -174dBm/Hz remains thecurrently used value.

Fig 15: Reciprocal mixing.


138

17References

It is easy to find tens of references on theInternet, professional and amateur topics.Here is an extract of those I read andassimilated. [1] A Simple Model of Feedback Oscilla-tor Noise Spectrum, D.B. Leeson, Pro-ceedings of the IEEE, February 1966[2] Oscillator Phase Noise: Theory andPrediction, K.V. Puglia, Microwave Jour-nal, 9/2007[3] All About Noise in Oscillators, UlrichL. Rohde, KA2WU/DJ2LR/HB9AWE,QEX 12/1993, 1/1994, 2/1994[4] Oscillators: A New Look at an OldModel, Stan Alechno, Microwave & RF,12/2001, 1/2003[5] Comprendre les performances desrécepteurs radio, F4BUC, Radio-REFjuin 2003[6] Phase Noise :Theory versus Practical-ity, John Esterline, Microwave & RF,April 2008 [7] Le bruit de phase, F5RCT, REF 67,3/2007[8] Difficultés de pilotage par OCXO enSHF, F9HX, Radio-REF 12/2004[9] Un pilote VHF stabilisé par PLL pourvotre transverter SHF, F9HX, Radio-REF11/2003[10] Millimetre-Wave LO References &Phase Considerations, WA1ZMS, Micro-wave Update 2004 [11] Le bruit des régulateurs de tensionscontinues, F9HX, Radio-REF 3/2008[12] 2 m Direct Conversion Transceiver,F9HX,VHF-Communications 1/2003

[13] Low-Noise VHF-Oscillator withDiode Tuning, DJ7VY, VHF Communi-cations 2/81

[14] Theory of Intermodulation and Re-ciprocal Mixing: Practice, Definitionsand Measurements in Devices and Sys-t e m s , U l r i c h L . R o h d eKA2WEU/DJ2LR/HB9AWE, Part 1,QEX 11-12/2002 [15] Local Oscillator Phase Noise and itsEffect on Receiver Performance, C. JohnGrebenkemper, Watkins-Johnson Com-pany, Tech-notes 11/12/81[16] Noise Reference, Applied RadioLabs, http://www.radiolab.com.au[17] The Receiver Noise Equation: AMethod for System Level Design on aRF Receiver, Hyung JounYoo & Ji-HoonKim, Microwave Journal, 8/2002[18] Additive (Residual) Phase NoiseMeasurements of Amplifiers, FrequencyDividers and Frequency Multipliers, Ja-son Breitbarth & Joe Koebel, MicrowaveJournal, June 2008


139

Documents

The harmful effects of local oscillator noisemillimeterwave.free.fr/F9HX lo noise.pdfTo test phase modulation, an 8038 deliv-Fig 5: Spectral density of a very low noise OCXO spoiled