Put two radio systems on the same structure and they stop being independent. The transmitter on one antenna pours energy into the receiver on the next antenna along, and without enough separation the receiver goes deaf.
So how far apart do the antennas need to be? This guide answers that properly, starting with the numbers.
The Short Answer
Antenna separation is not really about distance. It is about isolation, the loss in decibels between one antenna’s port and another.
You need enough isolation to keep two things below the receiver’s limits at once: the transmitter’s broadband noise that lands in the receive channel, and the transmitter’s own carrier reaching the receiver front end. The larger of those two requirements sets the target.
Separation buys isolation, and the standard first-order estimates come from ITU-R Report M.2244:
Side by side (horizontal): I = 22 + 20·log₁₀(d / λ) − G_tx − G_rx (dB)
Stacked (vertical): I = 28 + 40·log₁₀(d / λ) (dB)
Here d is the separation in metres, λ is the wavelength in metres, and gains are in dBi.
Key takeaway: The vertical form grows at 40 dB per decade of separation, the horizontal form at only 20 dB. That is why co-sited antennas are stacked one above the other rather than spread out sideways.
A Worked Example First
Formulas make more sense once you have seen them produce a number, so here is the whole method on one problem.
Two UHF land-mobile systems have to share a rooftop at 450 MHz, where the wavelength is 0.67 metres.
The interferer. The transmitter runs 25 watts, which is 44 dBm.
The victim. The receiver uses a 12.5 kHz channel. Its thermal noise is minus 174 plus 10·log₁₀(12500), or about minus 133 dBm, and with a 6 dB noise figure its noise floor sits at minus 127 dBm. The operator will accept 1 dB of desense, which allows interference about 6 dB below that floor, near minus 133 dBm.
The requirement. Suppose the transmitter’s broadband noise landing in the receive channel measures minus 63 dBm at its antenna port. The isolation needed is minus 63 minus minus 133, which is 70 dB. A blocking check agrees: if the receiver overloads at minus 26 dBm, the requirement is 44 minus minus 26, again 70 dB.
The separation. Solve the vertical formula for 70 dB. Setting 28 + 40·log₁₀(d / λ) equal to 70 gives d / λ of about 11.2, so the separation is 11.2 times 0.67 metres, or about 7.5 metres of vertical stacking. That is a realistic mast.
Trying the same job side by side, with two modest 5 dBi antennas, would need more than five hundred metres of horizontal separation. No rooftop provides it.
Key takeaway: 70 dB of isolation is 7.5 metres when the antennas are stacked and several hundred metres when they are side by side. Stack them.
The sections below unpack each step.
What Isolation Means
Antenna-to-antenna isolation is the ratio, in decibels, of the power delivered into a transmitting antenna to the power that reaches the port of a nearby receiving antenna.
If a transmitter puts 44 dBm into its antenna and 70 dB of isolation separates the two, then minus 26 dBm arrives at the receiver’s port before any filtering. Every interference mechanism is decided by comparing a power level against a receiver threshold, and isolation is the term that stands between them.
The Three Ways Co-Located Antennas Interfere
A proper check has to consider all three, because any one can be the limiting factor.
Transmitter noise and desensitisation
Every transmitter emits more than its wanted carrier. A real power amplifier produces a broad skirt of noise, and some of it lands inside the neighbouring receiver’s passband.
There it adds to the receiver’s own thermal noise and raises the effective noise floor. A weaker wanted signal is then lost in the grass, so usable sensitivity degrades. This is receiver desensitisation. The underlying receiver theory is set out in our companion article on the noise floor, receiver sensitivity and SNR.
Blocking
The second mechanism is the transmitter’s carrier itself. Even far off frequency, a strong nearby carrier can drive the receiver’s low-noise amplifier and first mixer into compression, an effect called blocking or overload.
The receiver’s gain sags, its noise figure worsens, and reciprocal mixing with the local oscillator’s phase noise can smear the carrier across the wanted channel. Blocking depends on the raw carrier power reaching the receiver, not on how much falls in band, so it is a separate calculation from desense.
Intermodulation
Once two or more carriers are present, mixing in any non-linear element generates intermodulation products at new frequencies, and a third-order product can land on a receive channel.
The non-linearity can sit in a transmitter’s output stage, a receiver’s front end, or a corroded joint on the tower. We cover the active case in intermodulation, IM3 and IP3 and the passive case in passive intermodulation, PIM, and how to prevent it. More isolation lowers the offending carriers at the non-linear element, so it reduces the products.
Turning Receiver Tolerance Into a dB Budget
The isolation target comes from the receiver, not from a rule of thumb.
The desense limit
Decide how much degradation you will accept. A degradation of Δ dB means the interference sits at 10·log₁₀(10^(Δ/10) − 1) decibels relative to the noise floor. That produces three numbers worth memorising:
3 dB desense allows interference equal to the noise floor.
1 dB desense allows interference about 6 dB below the noise floor.
0.5 dB desense allows interference about 9 dB below the noise floor.
The required isolation is the transmitter noise in the receive channel, measured at its port, minus that allowable level.
The blocking limit
This one is simpler. It is the transmitter carrier power minus the receiver’s stated blocking level. A 44 dBm carrier and a minus 26 dBm blocking level need 70 dB, independent of any noise consideration.
Key takeaway: Compute both limits and take the larger. For co-located land-mobile systems the governing figure commonly lands between 60 and 90 dB, but it is always the output of this budget, never an assumed number.
Both formulas come from ITU-R Report M.2244 and are also given in ITU-R Recommendation SM.337-6. The two geometries behave very differently.
The two ways to separate co-sited antennas. Side by side, the main beams face each other and the isolation grows at 20 dB for every tenfold increase in spacing. Stacked, the antennas present their nulls to each other, the gains drop out, and the isolation grows twice as fast at 40 dB per decade, which is why stacking is far more effective.
Side by side
The horizontal form is the Friis free-space relation written in wavelengths, so it grows at 20 dB per decade. The gains subtract because two antennas placed side by side point their main beams at each other.
Stacked
Two things change when you stack the antennas, and both help.
The slope doubles to 40 dB per decade, because stacked antennas couple through the near-field region where the field falls faster than in the far field.
The gains also disappear. A vertically polarised collinear antenna radiates towards the horizon and puts a deep null straight up and down, so a stacked pair presents its nulls to itself. M.2244 is explicit that the vertical form needs no gain information at all.
The difference in one table
The two slopes are the whole story. The table gives the isolation each form delivers as a function of separation in wavelengths, with the horizontal figures shown for zero-gain antennas so the geometry is compared like for like. Real gains only make the horizontal case worse.
Separation (d / λ)
Vertical
Horizontal (0 dBi)
10
68 dB
42 dB
20
80 dB
48 dB
50
96 dB
56 dB
100
108 dB
62 dB
Stacking reaches 68 dB at ten wavelengths. Reaching the same figure side by side takes roughly two hundred wavelengths, twenty times the distance.
The same isolation in metres
Because everything scales with wavelength, ten wavelengths is a very different distance across the bands. The table shows the vertical separation needed to reach ten wavelengths, which is where about 68 dB becomes available.
Band
Wavelength
10 λ vertical separation
150 MHz (VHF)
2.00 m
20.0 m
450 MHz (UHF)
0.67 m
6.7 m
900 MHz
0.33 m
3.3 m
1800 MHz
0.17 m
1.7 m
This is why VHF co-siting is so much harder than UHF. The wavelengths are large, so the same isolation demands a taller mast.
Where the Formulas Stop Working
These are planning-grade screening tools with real boundaries, and they return confident-looking numbers well outside the region where they hold.
The vertical form is only valid beyond ten wavelengths. M.2244 states this directly: equation (4) is applicable “when d_v is greater than 10λ”. Below that the near-field terms it assumes are negligible are no longer negligible, and the error takes no consistent sign.
The horizontal form assumes far-field spacing, d ≥ 2D²/λ for antennas of largest dimension D, which the report puts at about ten wavelengths for the simple dipole case. Its accuracy falls as antenna gain drops.
Two cautions apply even inside those bounds.
First, the vertical formula assumes vertically polarised collinear antennas that present nulls to each other. A panel or sector antenna does not, and the formula then overstates the isolation.
Second, orientation dominates. M.2244’s measurements were taken while rotating the antennas and applying electrical downtilt, and the isolation at a fixed separation can swing by roughly 20 dB from those effects alone. A single number from a single formula hides that spread.
Key takeaway: Use the closed form to size a design, then verify the tight pairs against real antenna patterns or a measurement before committing.
Our 3D antenna separation and co-siting tool computes two independent estimates for every pair, the M.2244 closed form and a pattern-aware Friis calculation that accounts for how the antennas actually point, and flags any pair where the two disagree or the geometry falls outside the bounds. A large disagreement is the finding.
Closing the Gap With Filtering
If the structure cannot provide the separation the budget demands, you buy the rest with hardware.
Isolation from separation and isolation from filtering add directly in decibels. A duplexer, cavity filter or band-reject filter in the transmit path, the receive path, or both, makes up the shortfall. If the geometry delivers 55 dB and the budget needs 70, you need 15 dB of extra filter isolation.
Tight co-sited sites are engineered exactly this way: reasonable physical separation to do the bulk of the work, then filtering to add the last stubborn margin where a taller mast is not an option.
A Note on Motorola R56
Co-siting separations are often quoted from Motorola’s R56 rules of thumb, such as 3 metres for 45 dB. Those are useful field heuristics, but they are frequency independent, and that is the catch.
Their implied slope is about 33 dB per decade rather than the 40 dB per decade of the vertical form, and their absolute values only line up with the physics near 266 MHz. The wavelength-based M.2244 formulas are correct across every band, so treat R56 as a sanity check rather than the source of truth.
Frequently Asked Questions
How much separation do I need between two antennas?
There is no fixed distance, because separation is only a means to an end. Work out the isolation your receiver needs in decibels, then solve the vertical formula for the separation that provides it. As an anchor, ten wavelengths of vertical stacking gives about 68 dB, which is roughly 6.7 metres at 450 MHz and a full 20 metres at 150 MHz.
Should I stack antennas vertically or space them horizontally?
Stack them vertically. Vertical separation buys isolation at 40 dB per decade against 20 dB per decade side by side, and it avoids the penalty of the antenna gains pointing at each other. Reaching a given isolation side by side typically needs about twenty times the distance.
What is receiver desensitisation?
It is the loss of receiver sensitivity caused by unwanted energy, most often a nearby transmitter’s broadband noise, raising the receiver’s effective noise floor. The wanted signal then competes with a higher noise level and coverage shrinks. One decibel of desense is a common design limit.
Can I trust the isolation a formula gives me?
Only inside its bounds, and only as a planning estimate. The M.2244 vertical form holds beyond ten wavelengths and assumes idealised antenna behaviour, and real isolation at a fixed separation can move by 20 dB or more with orientation and downtilt. Size the design with the formula, then verify the tight pairs by pattern analysis or measurement.
What isolation does a co-sited site need?
It depends on transmit power, frequency offset, transmitter noise, and the receiver’s sensitivity and blocking level. For co-located land-mobile systems the governing requirement commonly falls between 60 and 90 dB, but the only correct answer is the one your own budget produces.
Intermodulation is the interference created when two or more signals mix in a non-linear device and produce new signals at the sums and differences of their frequencies. The third-order products at 2f1 minus f2 and 2f2 minus f1 fall closest to the carriers and are the usual troublemakers. This guide explains active and passive intermodulation, where the products land, the third-order intercept point IP3 and the 3 to 1 rule, a worked multi carrier example, and the filtering, isolation and frequency planning that keep IM off a site.
RF interference is becoming more common as wireless systems multiply and spectrum becomes increasingly congested. Learn what causes it, why it is worsening, and what engineers can do to mitigate it.
Noise figure is how many decibels of noise a stage adds on top of a perfect receiver, and in a chain those figures do not add. The Friis cascade formula divides every later stage by the gain ahead of it, which is why the first stage sets the system noise figure and why a lossy feeder run in front of the amplifier is the most expensive mistake on the tower. This guide covers noise figure, noise factor and noise temperature, the cascade formula, why passive loss equals noise figure decibel for decibel, a worked four stage chain that gains 2.9 dB purely by moving the LNA, how much gain is enough, the dynamic range you pay for it, Y factor measurement, and when chasing a lower noise figure buys you nothing at all.
