Uplink and downlink LTE link budgets computed separately with the PRB aware uplink noise bandwidth, a cell edge service target converted to the required SINR through 3GPP TR 36.942, the shadow margin stated as a location probability, the limiting direction named rather than assumed, and every line labelled specification, input, assumption or derived.
Free forever on a Standard account. No credit card.
Every free LTE link budget on the internet computes the uplink noise floor over the channel bandwidth. That is wrong, and the error is large. One resource block is 12 subcarriers of 15 kHz, so 180 kHz exactly. A cell edge terminal is scheduled a handful of resource blocks, not the hundred a 20 MHz carrier spans, and the receiver integrates noise over the allocation alone. Four resource blocks is a 0.72 MHz noise bandwidth against 18 MHz for the full carrier, which is 14 dB of uplink budget thrown away. The error is pessimistic, which sounds safe right up until it is used to justify more sites than the network needs.
That correction has a consequence most tools never surface. At a wide uplink allocation the uplink limits, as everyone expects, because the terminal has a fraction of the base station power. Narrow the allocation and the uplink noise bandwidth falls with it, until somewhere in the single digits or tens of resource blocks the uplink overtakes the downlink and the DOWNLINK becomes the binding direction. A calculator using the channel bandwidth never sees that crossover and so recommends the wrong fix: more terminal power, when the problem is at the other end. This one prints the allocation at which the direction flips.
Nobody specifies a cell edge as minus one dB of SINR. They specify a service: so many kbit/s each way for the worst placed terminal. The tool takes that directly and derives the required SINR per direction through the link level model 3GPP itself publishes for the purpose, TR 36.942 Annex A.1, with the frame share and overhead of the actual resource grid, so the figure agrees with the LTE Throughput calculator to the digit. A target the allocation cannot carry is refused with the smallest allocation that can, and because the required SINR falls as the allocation widens while the noise rises, the allocation that gives the uplink the most is reported. The shadow fade margin can likewise be stated as a location probability at the cell edge against a sigma cited to ITU-R P.1546-6, so the design basis can say ninety per cent rather than eight dB.
The budget is built term by term with provenance on each line. UE maximum output power comes from TS 36.101 Table 6.2.2-1 for the selected band and power class, and where a band does not define that class the tool says so and flags the assumed figure. Base station power is read per port or as the total across ports, because datasheets are written both ways and the difference is three dB on two ports. Reference sensitivity from Table 7.3.1-1 is shown beside the derived figure as a sanity check, with the note that exact agreement would be a coincidence rather than a validation: one is a conformance minimum at a defined reference measurement channel, the other is kTB plus a noise figure plus a required SINR.
Maximum allowable path loss is inverted to a range through the Okumura Hata and COST 231 kernel from the Coverage Predictor, which agrees with an independent implementation to nine decimal places across 193 cases, and is applied only inside the domain it was fitted over: 150 to 2000 MHz, a base antenna of 30 to 200 m and a terminal of 1 to 10 m. At 3.5 GHz or on a 6 m pole no range is reported and the reason is stated; a result under 1 km or over 20 km is reported but flagged as extrapolated. One click carries the limiting MAPL to the Cell Range tool, whose TR 38.901 models cover what Hata cannot.
The term most calculators get wrong. Noise is integrated over the allocated resource blocks at 180 kHz each, not over the channel. The tool shows the allocation, the resulting noise bandwidth, and exactly how much budget a channel bandwidth calculation would have discarded, as its own headline figure.
Uplink and downlink are computed separately and compared. The answer is not assumed to be the uplink, because at narrow allocations it is not, and the recommended fix differs on either side of the crossover. The allocation at which the direction flips is printed in the summary.
Type the kbit/s the worst placed terminal must carry in each direction and the required SINR is derived through TR 36.942 Annex A.1, labelled derived with the clause. A target the allocation cannot carry is refused with the codeset ceiling and the smallest allocation that carries it. A manual SINR remains available for a measured vendor figure.
Widening the uplink allocation lowers the SINR the rate needs and raises the noise the receiver integrates, so the uplink MAPL peaks in between. With a service target the tool sweeps every allocation and reports the one that gives the uplink the most, and by how many dB against yours.
Pick the share of cell edge locations the design must serve and the margin is sigma times the normal quantile, with sigma cited to ITU-R P.1546-6 Annex 5 section 12 for the environment or typed. The limiting MAPL carries the probability it was built for, which is the sentence that goes in a design basis. Manual margin entry is retained.
Datasheets write base station power both ways. The tool asks which, adds 10 log of the ports for per port power and nothing for total, and the Ports combined line says what it did. Reading one as the other is 3 dB on two ports and 6 dB on four.
Not the better one. Quoting the range of the direction that happens to be stronger is how a coverage claim survives a design review and fails in the field.
Maximum output power comes from TS 36.101 Table 6.2.2-1 for the selected band and power class. Where a band does not define that class, the tool says so and flags the assumed figure instead of quietly substituting a typical value.
Table 7.3.1-1 sensitivity is displayed beside the derived figure, with the caveat that these are different quantities and exact agreement would be coincidence. It is a sanity check, not a validation, and the page says which.
Specification, input, assumption or derived, with the clause where there is one. A budget whose terms cannot be told apart is a budget whose errors cannot be found, and the assumptions that matter are named on the page.
Shadow fade, building entry loss, body loss and interference margin are separate lines. A single combined margin hides which allowance is doing the work and makes it impossible to argue with any one of them.
Hata and COST 231 were fitted from 150 to 2000 MHz with the base antenna between 30 and 200 m. Outside that, no range is reported and the reason is given; a short or long result is flagged as extrapolated. A model that quietly extrapolates is worse than one that declines, because the output looks identical either way, and 3.5 GHz is exactly where private LTE lives.
The limiting MAPL, its frequency, both antenna heights and the clutter class are carried to the LTE Cell Range and Coverage Estimator, where the TR 38.901 urban macro, street canyon and rural macro models take over. Its indoor option is turned off in the hand-off because the budget already carries its building entry loss.
A receiver integrates noise over what it is listening to. A cell edge terminal is scheduled a handful of resource blocks at 180 kHz each, not the whole carrier. Using the channel bandwidth overstates the uplink noise floor by 10 log10 of the ratio, which on a 20 MHz carrier with four resource blocks is 14 dB. The error is pessimistic, so it rarely gets challenged, and it is used to justify sites the network does not need.
No, and that is one of the most useful things this tool surfaces. At a wide uplink allocation the uplink limits, because the terminal has a fraction of the base station power. Narrow the allocation and the uplink noise bandwidth falls with it, and somewhere in the single digits or tens of resource blocks the downlink takes over. The tool prints the allocation at which that happens. A calculator that assumes the uplink always binds will recommend more terminal power for a problem that is at the other end of the link.
3GPP publishes the CQI and MCS tables but not the link level mapping from SINR to either, so any table that says a given MCS needs a given SINR is an invention. What 3GPP does publish, for exactly this purpose, is the attenuated Shannon bound of TR 36.942 Annex A.1 with its fitted parameters. The tool takes the kbit/s over the allocation, applies the frame share and overhead of the actual resource grid, and inverts that curve. The result is labelled derived with the clause, the curve's basis (1:2 antennas, Typical Urban channel) is printed under Sources, and if you have a measured vendor figure the manual SINR mode takes it instead.
Shadow fading is log normal around the median path loss, so a median budget serves half the locations at the edge. The margin that serves 90 per cent is sigma times 1.28, and 95 per cent is sigma times 1.64. Sigma comes from ITU-R P.1546-6 Annex 5 section 12 by service area, 8 dB urban, 10 dB suburban and 12 dB open, or you can type your own. The term in the budget shows the arithmetic and the limiting MAPL says which probability it carries. It is an edge probability; the area probability is higher.
Because equipment datasheets are written both ways, 2 x 20 W and 40 W both appear, and reading one as the other is a 3 dB error on two ports and 6 dB on four. Three dB of downlink budget is roughly a fifth of the range. The Ports combined line adds 10 log of the ports for per port power, adds nothing for total, and says which it did.
Because they are different quantities, and the page says so. Table 7.3.1-1 is a conformance minimum measured at a defined reference measurement channel, with implementation margin built into it. The derived figure is kTB plus a noise figure plus the SINR you asked for. They land close for good reasons, but exact agreement would be a coincidence rather than a validation, and treating it as one would be misleading.
They are labelled on the line. The receive diversity combining gain depends on branch correlation, which depends on the installation. A UE power class the band does not define is budgeted at 23 dBm and flagged. When a service target is converted to an SINR, three PDCCH symbols are assumed for the overhead and the curve is a generic 3GPP one rather than this vendor's equipment. Everything else is either a specification value with its clause, one of your inputs, or derived from the others.
Because the frequency or an antenna height is outside the domain Hata and COST 231 were fitted over, 150 to 2000 MHz and a base antenna of 30 to 200 m, or because the budget does not close anywhere from 20 m to 100 km. Extrapolating past that produces a number that looks exactly like a valid one, with nothing to indicate it is unsupported. Declining is the honest output, the tool says which bound was crossed, and the Cell Range tool takes the same MAPL to the TR 38.901 models that do cover 3.5 GHz and small cells.
The limiting MAPL, the frequency of the direction that produced it, both antenna heights and the clutter class. It turns Cell Range's indoor option off, because the budget already contains whatever building entry loss you put in it and applying P.2109 on top would count it twice; the notice on the far side says how many dB were already inside the MAPL. Cell Range then picks the TR 38.901 deployment from the clutter class and base height, and you can change it.
The Okumura Hata and COST 231 kernel is shared with the Coverage Predictor and agrees with an independent third party implementation to nine decimal places across 193 cases. It is the same kernel, not a second copy, which matters: a duplicated propagation model that drifts from the validated one is worse than not having the feature.
No. The range comes from a measurement fitted clutter model, and nothing in it sees a hill. For a first pass on cell count that is the right level of effort, and for anything past scoping the Coverage Predictor does the terrain work properly. The result says so rather than leaving it to be assumed.
Free forever on a Standard account. No credit card.
