It is a huge difference matching 2-2.5 GHz relative 300-900 MHz, using passive reactive components.
2.5 GHz is 25% above 2 GHz and 900 MHz is 300% above 300 MHz. It is a factor more then 10x in tuning range. Add that actual curve is way off from matching impedance goal.
It is also a huge diffrence if matching with real and lossy components, using real values, or ideal components and any non standard value .
Auto matching for finding best values and best topology using Murata LQW15 inductors result in this impedance:
Blue curve is vector sum of S11 and S22. Smith center is 50 Ohm.
Return loss and matching network values. It is a non symmetrical network but it works just as well if ground is replaced by signal.
Orange is unmatched and yellow is after matching.
If alternatively running auto-matching for 300-900 MHz is result as expected almost without value. Depending on what parameters that is most important can it be better to not tune at all or do a resistor match.
Resulting auto-matching, componet types, values and topology automatically optimized.:
Blue is unmatched and Yellow matched.
As can be seen is it not at all same components values and not same optimized topology for these both frequency ranges, which makes it very complex to find design a network that reduces losses for both these frequency ranges.
I do often design embedded wideband antennas for 4G and similar. A typical such wide freq range antenna is covering 700-2700 MHz with decent efficiency. A tuning network is always needed for these small antennas and also often small ground plane but the trick is often to mis-design the antenna impedance such that both lower and higher frequencies have gain of same matching network.
When actual mesurement for a frequency range creates a complete wide circle in Smith chart will tuning at one frequency range push away corresponding frequency at opposite side of Smith chart.
It is possible to match by chopping needed frequencies in smaller bands but also that is complicated if low loss is more important then a formal impedance match. If a matching topology chain becomes long do it tends to increase resistive losses so that matching seems to be improved but added component does not reduce loss total loss.
A resistive network can even be a better choice the using reactive components if matching is improved at similar amount of loss.
A complex topology with many components in a chain is very time consuming to develop due to several real world losses, non ideal ground and if distance between first and last components have costed phase delay and stray losses must often values be fine adjusted by trial and error.
There are simple methods to develop at least chains as above, with 4 components, in a rather well predicted way but but for actual curve is it of limited value.
A bit different way to do impedance matching is to do dynamic tuning. A wide frequency range can be covered by just two dynamic tuned components. I did use it for some cellphone antennas long ago, for embedded FM antenna 88-108 MHz as well as UHF (DVB-h). By using a high-Q antenna dynamic tuned could whole FM-band be covered with an antenna just 1 MHz wide in its 3db tuning range. Actual antenna was real small, 10x4x4 mm, but did still fullfil then existing efficiency specification, MoBile Radio Access Interface specification (MBRAI).